# Xiaomi MiMo API Open Platform

> Xiaomi MiMo API Open Platform provides high-performance inference services for Xiaomi's AI models, compatible with OpenAI and Anthropic API formats.

This platform offers comprehensive API documentation, integration guides, and detailed update logs for Xiaomi-related AI models. It is designed to empower developers to build and deploy next-generation intelligent applications and agents with ease.



--- DOCUMENT: First API Call ---
URL: https://mimo.mi.com/static/docs/quick-start/summary/first-api-call.md

# First API Call

## Supported API Types

{/* feishu-style:text-align:left */}
Xiaomi MiMo API Open Platform is compatible with OpenAI API and Anthropic API formats. You can use existing SDKs to access model inference services.

## Preparation Before Calling

### Log in to Xiaomi MiMo API Open Platform

{/* feishu-style:text-align:left */}
Currently, the platform only provides personal account login. You need to use a Xiaomi account to log in. If you already have a Xiaomi account, you can log in directly. If you don't have a Xiaomi account, you can visit the [Console](https://platform.xiaomimimo.com/#/console/usage) to register, or register in advance at [id.mi.com](https://id.mi.com/).

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 12.61%" />
<col style="width: 16.84%" />
<col style="width: 8.82%" />
<col style="width: 61.730000000000004%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th colspan="2"><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td rowspan="2"><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Real-time Inference</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li><li>Anthropic Compatibility Protocol: `https://api.xiaomimimo.com/anthropic`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Batch Inference</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://batch-api-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [**Batch Inference**](https://platform.xiaomimimo.com/console/batch) to get your dedicated Base URL.</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: center;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li><li>Anthropic Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/anthropic`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

> Please keep your API Key safe to avoid leakage that may result in quota theft. It is recommended to configure the API Key in environment variables.

## Quick Integration Examples

{/* feishu-style:text-align:left */}
You can copy the following API example code and replace the API Key value to quickly make calls. If you use the Token Plan, you need to replace the BASE_URL and use the dedicated API Key.

{/* feishu-style:text-align:left */}
The following system prompts are highly recommended, please choose from English and Chinese version.

> Chinese version
>
> ```json
> 你是MiMo（中文名称也是MiMo），是小米公司研发的AI智能助手。
> 今天的日期：{date} {week}，你的知识截止日期是2024年12月。
> ```

> English version
>
> ```json
> You are MiMo, an AI assistant developed by Xiaomi.
> Today's date: {date} {week}. Your knowledge cutoff date is December 2024.
> ```

### OpenAI Chat Completions API Compatibility

{/* feishu-style:text-align:left */}
MiMo models are compatible with the OpenAI Chat Completions API and support invocation via the Python SDK and Curl. Integration examples are provided below.

<Tab>
  <TabItem label={`Python SDK`}>

{/* feishu-style:text-align:left */}
1. Install the OpenAI Python SDK by running the following command:

```shell
# If the run fails, you can replace pip with pip3 and run again
pip install -U openai
```

{/* feishu-style:text-align:left */}
2. Call the API:

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "please introduce yourself"
        }
    ],
    max_completion_tokens=1024,
    temperature=1.0,
    top_p=0.95,
    stream=False,
    stop=None,
    frequency_penalty=0,
    presence_penalty=0
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "please introduce yourself"
        }
    ],
    "max_completion_tokens": 1024,
    "temperature": 1.0,
    "top_p": 0.95,
    "stream": false,
    "stop": null,
    "frequency_penalty": 0,
    "presence_penalty": 0
}'
```

  </TabItem>
</Tab>

### Anthropic Messages API Compatibility

{/* feishu-style:text-align:left */}
MiMo models are compatible with the Anthropic Messages API and support invocation via the Python SDK and Curl. Integration examples are provided below.

<Tab>
  <TabItem label={`Python SDK`}>

{/* feishu-style:text-align:left */}
1. Install the Anthropic Python SDK by running the following command:

```shell
# If the run fails, you can replace pip with pip3 and run again
pip install -U anthropic
```

{/* feishu-style:text-align:left */}
2. Call the API:

```python
import os
from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/anthropic"
)

message = client.messages.create(
    model="mimo-v2.6-pro",
    max_tokens=1024,
    system="You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "please introduce yourself"
                }
            ]
        }
    ],
    top_p=0.95,
    stream=False,
    temperature=1.0,
    stop_sequences=None
)

print(message.content)
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/anthropic/v1/messages' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "max_tokens": 1024,
    "system": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.",
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "please introduce yourself"
                }
            ]
        }
    ],
    "top_p": 0.95,
    "stream": false,
    "temperature": 1.0,
    "stop_sequences": null
}'
```

  </TabItem>
</Tab>

### Make Multi-turn Tool Calls in Thinking Mode

{/* feishu-style:text-align:left */}
During the multi-turn tool calls process in thinking mode, the model returns a `reasoning_content` field alongside `tool_calls`. To continue the conversation, it is recommended to keep all previous `reasoning_content` in the `messages` array for each subsequent request to achieve the best performance.

{/* feishu-style:text-align:left */}
The requested example is as follows:

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "messages": [
        {
            "role": "assistant",
            "content": "Hello! I am MiMo.",
            "reasoning_content": "Okay, the user just asked me to introduce myself. That is a pretty straightforward request, but I should think about why they are asking this."
        },
        {
            "role": "user",
            "content": "What is the weather like in Hebei?"
        }
    ],
    "model": "mimo-v2.6-pro",
    "max_completion_tokens": 1024,
    "temperature": 1.0,
    "stream": false,
    "tools": [
        {
            "type": "function",
            "function": {
                "name": "get_current_weather",
                "description": "Get the current weather in a given location",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "location": {
                            "type": "string",
                            "description": "The city and state, e.g. San Francisco, CA"
                        },
                        "unit": {
                            "type": "string",
                            "enum": [
                                "celsius",
                                "fahrenheit"
                            ]
                        }
                    },
                    "required": [
                        "location"
                    ]
                }
            }
        }
    ],
    "tool_choice": "auto"
}'
```

## Check Usage Information

{/* feishu-style:text-align:left */}
On the [Usage Information](https://platform.xiaomimimo.com/#/console/usage) page, you can view and export detailed data of your account's model Token usage and request counts by date.


--- DOCUMENT: Models ---
URL: https://mimo.mi.com/static/docs/quick-start/summary/model.md

# Models

{/* feishu-style:text-align:left */}
This page lists all the models currently supported by the Xiaomi MiMo API Open Platform, including model capabilities, length limits, and rate-limiting quotas, to help you select the appropriate model based on your usage scenario. 

### Rate Limiting Instructions

{/* feishu-style:text-align:left */}
The platform sets a model concurrency limit for each account. When the server load is high, response delays or ` 429 ` error may occur. We recommend that you reasonably plan your request frequency and implement request retry and backoff strategies in high-concurrency scenarios to avoid triggering rate limits. 

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

- **RPM (Requests Per Minute)** : The maximum number of requests initiated per minute. The calculation scope is the sum of the total number of requests from all API Keys under a single account when calling the same model.

- **TPM (Tokens Per Minute)** : The maximum number of Tokens that can be interacted with per minute. The calculation scope is the sum of the total number of requested Tokens for all API Keys under a single account when calling the same model.

</div>

{/* feishu-style:text-align:left */}
<br></br>

### Text Generation Model

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

- `mimo-v2.5-pro` and `mimo-v2.5` **will be officially deprecated at 10:00 (Beijing time) on October 21,2026. It is recommended to switch to the new version of the models as soon as possible. For details, please refer to** [**Model Deprecation**](https://mimo.mi.com/docs/zh-CN/updates/deprecate)**.**

</div>

<table>
<colgroup>
<col style="width: 29.2%" />
<col style="width: 27.53%" />
<col style="width: 24.67%" />
<col style="width: 18.599999999999994%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**模型 ID (Model ID)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**能力支持**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**长度限制（token）** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**限流**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`</span><br /><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`</span><br /><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5`(to be deprecated)</span></td>
<td rowspan="2"><ul><li>**Full-modal Understanding**</li><li>Text Generation</li><li>Deep Thinking</li><li>Streaming Output</li><li>Function Call</li><li>Structured Output</li><li>Web Search</li></ul></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Context Window: 1M</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Maximum Output: 128K</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Maximum RPM: 100</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Maximum TPM: 10M</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro-ultraspeed`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Context Window: 1M</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Maximum Output: 128K</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Customized services available, please [contact us](https://platform.xiaomimimo.com/contact?userId=2451887661)</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-pro`(to be deprecated)</span></td>
<td><ul><li>Text Generation</li><li>Full-modal Understanding</li><li>Deep Thinking</li><li>Streaming Output</li><li>Function Call</li><li>Structured Output</li><li>Web Search</li></ul></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Context Window: 1M</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Maximum Output: 128K</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Maximum RPM: 100</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Maximum TPM: 10M</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
<br></br>

### Automatic Speech Recognition (ASR) Model 

<table>
<colgroup>
<col style="width: 30.04%" />
<col style="width: 26.7%" />
<col style="width: 24.67%" />
<col style="width: 18.590000000000003%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Model ID**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Capability Support**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Length Limit (token)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Rate Limiting**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-asr`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Speech Recognition</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Context Window: 8k</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Maximum Output: 2k </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Maximum RPM: 100</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Maximum TPM: 10k</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
<br></br>

### Text-to-Speech (TTS) Model

<table>
<colgroup>
<col style="width: 30.04%" />
<col style="width: 26.7%" />
<col style="width: 24.67%" />
<col style="width: 18.590000000000003%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Model ID (Model ID)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Capability Support**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Length Limit (token)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Rate Limiting**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Speech Synthesis</span></td>
<td rowspan="3"><span style="display: inline-block; width: 100%; text-align: left;">Context Window: 8K</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Maximum Output: 8K </span></td>
<td rowspan="3"><span style="display: inline-block; width: 100%; text-align: left;">Maximum RPM: 100</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Maximum TPM: 10M</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts-voiceclone`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Speech Synthesis</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Timbre Cloning</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts-voicedesign`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Speech Synthesis</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Timbre Design</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
<br></br>

### Quick Selection Guide

<table>
<colgroup>
<col style="width: 63.03%" />
<col style="width: 36.97%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Requirement Scenario</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Recommendation Model</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Complex projects, long-term tasks, high-value work, cybersecurity and scientific research requirements</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Frequent Calls and Large-Scale Tasks in Professional Office Scenarios</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Production scenarios with strong real-time interaction and sensitivity to response speed</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro-ultraspeed`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Speech-to-Text (supports Chinese and English)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-asr`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Text-to-Speech (Standard Preset Voice)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Voice Cloning (Upload Audio Sample)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts-voiceclone`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Custom Timbre Design</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts-voicedesign`</span></td>
</tr>
</tbody>
</table>


--- DOCUMENT: Web Search ---
URL: https://mimo.mi.com/static/docs/quick-start/usage-guide/text-generation/tool-calling/web-search.md

# Web Search

{/* feishu-style:text-align:left */}
Web Search is a basic online search tool that helps your large model obtain real-time public online information (such as news, products, weather, etc.).

{/* feishu-style:text-align:left */}
**Core Capabilities**

- **Flexible search modes**: Supports forced search and intent recognition. With intent recognition enabled, the model will autonomously decide whether to perform an online search without manual triggering.

- **Early search source return**: In the streaming response, the first packet will return all search sources.

- **Hybrid multi-tool invocation**: Can work with custom functions and tools; the model will automatically determine invocation priority and necessity.

- **Flexible response modes**: Supports both streaming and non-streaming responses, and both methods will return search and summary content.

## Quick Start

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

The [Web Search Plugin](https://platform.xiaomimimo.com/#/console/plugin) must be activated prior to use.

</div>

### Enable the Service

1. Go to [Console → Plugin Management](https://platform.xiaomimimo.com/#/console/plugin), and activate the Web Search Plugin.

1. Web Search Plugin fee, refer to the [Pricing](https://platform.xiaomimimo.com/#/docs/pricing). Note: **Search invocation is determined by the model. A single search round (if required by the model) may initiate multiple keywords concurrently, resulting in multiple invocations of the Internet Content Plugin. You may use the** `max_keyword` **parameter to limit the maximum number of keywords per search round, thereby further controlling invocation frequency and costs**.

{/* feishu-style:text-align:left */}
**Note**：For preparations such as obtaining an API Key, please refer to [First API Call](https://platform.xiaomimimo.com/#/docs/quick-start/first-api-call).

### Sample Code

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "武汉明天天气怎么样？"
        }
    ],
    max_completion_tokens=1024,
    temperature=1.0,
    top_p=0.95,
    stream=False,
    stop=None,
    frequency_penalty=0,
    presence_penalty=0,
    extra_body={
        "thinking": {"type": "disabled"}
    },
    tools=[
        {
            "type": "web_search",
            "max_keyword": 3,
            "force_search": True,
            "limit": 1,
            "user_location": {
                "type": "approximate",
                "country": "China",
                "region": "Hubei",
                "city": "Wuhan"
            }
        }
    ],
    tool_choice="auto"
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "user",
            "content": "武汉明天天气怎么样？"
        }
    ],
    "tools": [
        {
          "type": "web_search",
          "max_keyword": 3,
          "force_search": true,
          "limit": 1,
          "user_location": {
            "type": "approximate",
            "country": "China",
            "region": "Hubei",
            "city": "Wuhan"
          }
        }
    ],
    "max_completion_tokens": 1024,
    "temperature": 1.0,
    "top_p": 0.95,
    "stream": false,
    "stop": null,
    "frequency_penalty": 0,
    "presence_penalty": 0,
    "thinking": {
        "type": "disabled"
    }
}'
```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Response**

```json
{
    "id": "9ffcbd3a-d3a2-4788-9edd-d8a79e47ae15_33e773cc83af482aa0b41b9fd120fdf1",
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "message": {
                "content": "武汉明天（9月23日）将以阴天为主，局部有阵雨，气温在22~25℃之间。受冷空气影响，近期气温有所回落，体感偏凉，出门建议携带雨具并适当添衣。",
                "role": "assistant",
                "annotations": [
                    {
                        "type": "url_citation",
                        "url": "https://www.weather.com.cn/weather/101200101.shtml",
                        "title": "武汉天气预报,武汉7天天气预报,武汉15天天气预报,武汉天气查询",
                        "summary": " 武汉天气预报，及时准确发布中央气象台天气信息，便捷查询武汉今日天气，武汉周末天气，武汉一周天气预报，武汉 …",
                        "site_name": "中国天气网",
                        "publish_time": "2026-09-21T08:32:25.0000000",
                        "logo_url": "https://weather.com.cn/favicon.ico"
                    },
                    {
                        "type": "url_citation",
                        "url": "https://www.qweather.com/weather/wuhan-101200101.html",
                        "title": "武汉市天气 空气质量 降水 预警以及武汉市历史天气 | 和风天气",
                        "summary": "2级 东北风56%相对湿度弱紫外线24°体感温度25km能见度0.0mm降水量1015hPa大气压24小时预报   5am    7am    8am    9am    11am    1pm    2pm    3pm    5pm    7pm    8pm    9pm    11pm    1am    2am    3am 温度 查看完整天气地图  未来预报今天09月15日 29°  22° 周三09月16日 26°  18° 周四09月17日 22°  18° 周五09月18日 24°  21° 周六09月19日 28°  23° 周日09月20日 31°  24° 周一09月21日 32°  22°   30天预报  空气质量优 34 PM2.5 39 PM10 92 O3 0.7 CO 11 SO2 17 NO2  预报和更多  太阳现在04:45现在04:45高度角-18°方位75°ENE月亮 蛾眉月 现在04:45现在04:45高度角-64°方位68°ENE生活指数舒适度指数较舒适洗车指数较适宜穿衣指数热感冒指数少发运动指数较不宜旅游指数适宜紫外线指数弱空气污染扩散指数良武汉市的历史天气  更多历史天气  武汉市气象概述武汉属亚热带季风气候区武汉稳定≥10℃活动积温在5000℃～5300℃之间,年无霜期240天。温度 查看完整天气地图  卫星云图天气排行  今日最高温    空气质量...",
                        "site_name": "和风天气",
                        "publish_time": "2026-09-15T00:00:00.0000000",
                        "logo_url": "https://www.qweather.com/favicon.ico"
                    },
                    {
                        "type": "url_citation",
                        "url": "https://www.sina.cn/news/article/comos_nishzsi2828295.html",
                        "title": "武汉未来一周天气预报:晴好升温至33℃,下周初阵雨来袭+FAQ|武汉天气|一周天气预报|武汉未来7天|武汉温度|武汉降雨_新浪新闻",
                        "summary": "和风天气30天预报指出,未来一周最高温出现在9月25日(33℃),最低温出现在9月... 🌡️ 最高33℃❄️ 最低21℃☀️ 晴好为主🌧️ 阵雨2天🍃 风力1-2级  | 日期 | 天气 | 气温(℃) | 风力 | 空气质量 |  |-----|-----|-----|-----|-----|  | 9月19日(周六) | 晴转多云 | 21~27 | 东北风2级 | 优 |   | 9月20日(周日) | 多云 | 22~30 | 北风1级 | 良 |   | 9月21日(周一) | 阴转阵雨 | 22~29 | 东北风2级 | 优 |   | 9月22日(周二) | 阴 | 22~25 | 东北风2级 | 优 |   | 9月23日(周三) | 多云 | 21~27 | 东北风1级 | 优 |   | 9月24日(周四) | 阴转多云 | 23~30 | 东北风1级 | 良 |   | 9月25日(周五) | 多云转中雨 | 24~33 | 北风1级 | 优| 　　数据来源:综合天气网、和风天气、全国天气网、2345天气预报(2026年9月19日发布) 02降雨分布:仅21日、25日有阵雨 ⏰ 出行提醒  9月21日(周一)阵雨伴雷电,建议错峰出行或备雨具  9月25日(周...",
                        "site_name": "新浪新闻",
                        "publish_time": "2026-09-19T06:13:50.0000000",
                        "logo_url": "https://sina.cn/favicon.ico"
                    }
                ],
                "tool_calls": null
            }
        }
    ],
    "created": 1790008349,
    "model": "mimo-v2.6-pro",
    "object": "chat.completion",
    "usage": {
        "completion_tokens": 60,
        "prompt_tokens": 1476,
        "total_tokens": 1536,
        "completion_tokens_details": {
            "reasoning_tokens": 0
        },
        "prompt_tokens_details": {
            "cached_tokens": 0
        },
        "web_search_usage": {
            "tool_usage": 3,
            "page_usage": 3
        }
    }
}
```

{/* feishu-style:text-align:left */}
For detailed parameters and invocation instructions, please refer to the [OpenAI Chat Completions API](https://platform.xiaomimimo.com/#/docs/api/text-generation/openai-api). Other API protocols are not supported for the time being.

## Supported models

{/* feishu-style:text-align:left */}
Currently, the `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro` and `mimo-v2.5` models are supported.

## Price

{/* feishu-style:text-align:left */}
The billing for the Web Search Plugin consists of the following two parts:

- **Usage of Web Search**: The number of times internet resources appear in one response of the Internet Service Plugin.

   - Cost per 1,000 calls from the Web Search tool: China ¥16 / 1K requests、Overseas $5 / 1K requests.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

When invoking Web Search via API, one search round will initiate concurrent keyword searches according to the `max_keyword` value, resulting in multiple uses of this plugin.

</div>

- **Model token fee**: The webpage content from internet search will be appended to the prompt, increasing the model’s input tokens. Billing is based on the model’s standard price. For price details, please refer to [Pricing and Rate Limits](https://platform.xiaomimimo.com/#/docs/pricing).

## FAQ

{/* feishu-style:text-align:left */}
**Why doesn’t the model perform a web search after enabling online search?** 

{/* feishu-style:text-align:left */}
There may be three reasons:

- **Cache**: There is a 5-minute cache period after enabling / disabling online search. The online search switch will not take effect immediately within 5 minutes.

- **Model determines no need for search**: The model judges that the current query does not involve real-time information and can be answered directly with its own knowledge. To force a search, set `force_search: true`.

- **Model not supported**: Currently, `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro` and `mimo-v2.5` support web search.


--- DOCUMENT: Deep Thinking ---
URL: https://mimo.mi.com/static/docs/quick-start/usage-guide/text-generation/deep-thinking.md

# Deep Thinking

{/* feishu-style:text-align:left */}
Deep Thinking enables the model to perform deep reasoning before generating the final answer, analyzing problems step-by-step through an internal Chain of Thought (CoT), significantly improving accuracy on complex tasks. It is suitable for scenarios requiring deep analysis, such as complex reasoning, code generation, mathematical computation, and multi-step analysis.

{/* feishu-style:text-align:left */}
**Core Capabilities**

- **Deep Reasoning**: Breaks down complex problems into multiple steps for analysis, improving reasoning accuracy

- **Transparent Thinking Process**: Returns the complete thinking process, enhancing interpretability

- **Flexible Control**: Supports enabling/disabling, allowing on-demand use based on task complexity

## Supported Models

{/* feishu-style:text-align:left */}
Currently, the `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro` and `mimo-v2.5` models are supported.

## Request Parameters

{/* feishu-style:text-align:left */}
Set the `thinking.type` parameter in the request to control deep thinking: `enabled` to enable deep thinking or `disabled` to disable deep thinking.

{/* feishu-style:text-align:left */}
**Default Status:** 

- Enabled by default: `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro` and `mimo-v2.5`

## Important Notes

### Parameter Limitations

{/* feishu-style:text-align:left */}
In deep thinking, `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro` and `mimo-v2.5` models do not support custom `temperature` and `top_p` parameters. Even if these parameters are provided, the actual effective values will be forced to use the recommended defaults of `1.0` and `0.95`.

### Multi-turn Conversation Pass-through Requirements

{/* feishu-style:text-align:left */}
When deep thinking is enabled in Agent product multi-turn conversations, and historical conversations contain tool calls, the assistant responses passed back in all subsequent user interaction rounds that contain tool calls <strong>must completely pass back the `reasoning_content` field, otherwise the API will return a 400 error</strong>. For the correct pass-back method, please refer to the "Multi-turn Tool Calls in Thinking Mode" section in the Call Examples.

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

If historical `reasoning_content` is missing, the model's context will be incomplete, which may result in decreased instruction following and increased hallucinations.

</div>

{/* feishu-style:text-align:left */}
**Affected Agent Products:** 

<table>
<colgroup>
<col style="width: 50%" />
<col style="width: 50%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Protocol</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Affected Agent Products</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">OpenAI Compatible Protocol</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">TRAE, Cursor, Roo Code, Codex, GitHub Copilot CLI, Zed, AutoGen, Goose</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Anthropic Compatible Protocol</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">TRAE, GitHub Copilot CLI, AutoGen, Goose, OpenClaw, OpenCode, Kilo Code</span></td>
</tr>
</tbody>
</table>

### Other Notes

1. **Output Length Limit**: `max_completion_tokens` limits the total length of thinking content and the final answer. If the thinking process is long, the token space available for the final answer will be reduced accordingly. It is recommended to set a sufficient `max_completion_tokens` to avoid answer truncation.

1. **Response Time**: Enabling deep thinking will increase response latency, especially for complex tasks. It is recommended to use `stream: true` to view the thinking process in real-time.

## Call Examples

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

The `thinking` field is not a standard OpenAI parameter. When passing thinking-related parameters via the OpenAI Python SDK, they must be included in `extra_body`.

</div>

### Thinking Enabled

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "Introduce machine learning in three sentences."
        }
    ],
    max_completion_tokens=1024,
    extra_body={
        "thinking": {"type": "enabled"}
    }
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "Introduce machine learning in three sentences."
        }
    ],
    "max_completion_tokens": 1024,
    "thinking": {
        "type": "enabled"
    }
}'
```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Response Example**

```json
{
    "id": "045bd1a4-688a-4e53-9de2-7631205e0182_5f467b0704184928936bafb270896022",
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "message": {
                "content": "Machine learning is a branch of artificial intelligence that enables computers to learn from data and improve their performance over time without being explicitly programmed for every task. It works by feeding algorithms large datasets, allowing them to identify patterns, make predictions, or take actions based on what they've learned. Common applications include spam filters, recommendation systems, image recognition, and language translation, making it a powerful tool that is reshaping industries from healthcare to finance.",
                "role": "assistant",
                "tool_calls": null,
                "reasoning_content": "The user wants a concise three-sentence introduction to machine learning."
            }
        }
    ],
    "created": 1790008551,
    "model": "mimo-v2.6-pro",
    "object": "chat.completion",
    "usage": {
        "completion_tokens": 103,
        "prompt_tokens": 60,
        "total_tokens": 163,
        "completion_tokens_details": {
            "reasoning_tokens": 14
        },
        "prompt_tokens_details": {
            "cached_tokens": 0
        }
    }
}
```

### Thinking Disabled

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "Write a short paragraph about the beauty of nature."
        }
    ],
    max_completion_tokens=1024,
    extra_body={
        "thinking": {"type": "disabled"}
    }
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "Write a short paragraph about the beauty of nature."
        }
    ],
    "max_completion_tokens": 1024,
    "thinking": {
        "type": "disabled"
    }
}'
```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Response Example**

```json
{
    "id": "ab231e1e-4d8c-4084-84bf-521959063597_b7c5e3eece0143ae98993a27e86e6df7",
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "message": {
                "content": "There is an enduring, quiet magic in the world outside our windows. It’s found in the delicate, painted canvas of a sunrise bleeding across the horizon, and in the intricate architecture of a single leaf, each vein a tiny river carrying life. The power of nature's beauty lies not just in its grandest vistas—the silent, snow-capped mountains and vast, churning oceans—but also in its most intimate details: the scent of rain on dry earth, the gentle rhythm of waves on a shore, and the resilience of a wildflower pushing through a crack in the stone. It is a constant, unfolding masterpiece that whispers to the soul, a powerful reminder of a force that is at once both magnificent and profoundly soothing.",
                "role": "assistant",
                "tool_calls": null
            }
        }
    ],
    "created": 1790008574,
    "model": "mimo-v2.6-pro",
    "object": "chat.completion",
    "usage": {
        "completion_tokens": 146,
        "prompt_tokens": 64,
        "total_tokens": 210,
        "completion_tokens_details": {
            "reasoning_tokens": 0
        },
        "prompt_tokens_details": {
            "cached_tokens": 0
        }
    }
}
```

### Streaming Response (Thinking Enabled)

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

During streaming responses, thinking content and answer content are output sequentially: first, the thinking process is returned step-by-step via `reasoning_content`, and after thinking is complete, the final answer is output step-by-step via `content`.

</div>

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "Give me some tips for improving work efficiency."
        }
    ],
    max_completion_tokens=1024,
    stream=True,
    extra_body={
        "thinking": {"type": "enabled"}
    }
)

for chunk in completion:
    print(chunk.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "Give me some tips for improving work efficiency."
        }
    ],
    "max_completion_tokens": 1024,
    "stream": true,
    "thinking": {
        "type": "enabled"
    }
}'
```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Response Example**

```json
data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":"","role":"assistant","tool_calls":null,"reasoning_content":null},"finish_reason":null,"index":0}],"created":1790008592,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":null,"role":null,"tool_calls":null,"reasoning_content":"The user is asking"},"finish_reason":null,"index":0}],"created":1790008592,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":null,"role":null,"tool_calls":null,"reasoning_content":" for tips to"},"finish_reason":null,"index":0}],"created":1790008592,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":null,"role":null,"tool_calls":null,"reasoning_content":" improve work efficiency."},"finish_reason":null,"index":0}],"created":1790008592,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

...

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":null,"role":null,"tool_calls":null,"reasoning_content":" tips"},"finish_reason":null,"index":0}],"created":1790008593,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":null,"role":null,"tool_calls":null,"reasoning_content":"."},"finish_reason":null,"index":0}],"created":1790008593,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":"# Tips","role":null,"tool_calls":null,"reasoning_content":null},"finish_reason":null,"index":0}],"created":1790008593,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":" for Improving Work","role":null,"tool_calls":null,"reasoning_content":null},"finish_reason":null,"index":0}],"created":1790008593,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

...

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":"? ","role":null,"tool_calls":null,"reasoning_content":null},"finish_reason":null,"index":0}],"created":1790008602,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":"😊","role":null,"tool_calls":null,"reasoning_content":null},"finish_reason":null,"index":0}],"created":1790008602,"model":"mimo-v2.6-pro","object":"chat.completion.chunk"}

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[{"delta":{"content":null,"role":null,"tool_calls":null,"reasoning_content":null},"finish_reason":"stop","index":0}],"created":1790008602,"model":"mimo-v2.6-pro","object":"chat.completion.chunk","usage":null}

data: {"id":"ac06ba44-79d8-419a-830f-8e443a15b962_f3425553986841e7bd6d1799bbb89be8","choices":[],"created":1790008602,"model":"mimo-v2.6-pro","object":"chat.completion.chunk","usage":{"completion_tokens":467,"prompt_tokens":61,"total_tokens":528,"completion_tokens_details":{"reasoning_tokens":29},"prompt_tokens_details":{"cached_tokens":0}}}

data: [DONE]

```

### Multi-turn Tool Calls in Thinking Mode

{/* feishu-style:text-align:left */}
In multi-turn conversations with deep thinking enabled, passing back `reasoning_content` when tool calls are involved ensures thinking continuity and improves model output quality.

```python
import os
import json
from openai import OpenAI

# Initialize client
client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

# Define tools
tools = [
    {
        "type": "function",
        "function": {
            "name": "get_current_weather",
            "description": "Get the current weather for a given city",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {"type": "string", "description": "City name, e.g. Beijing"},
                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}
                },
                "required": ["location"]
            }
        }
    },
    {
        "type": "function",
        "function": {
            "name": "get_time",
            "description": "Get the current time in a given timezone",
            "parameters": {
                "type": "object",
                "properties": {
                    "timezone": {"type": "string", "description": "Timezone, e.g. Asia/Shanghai"}
                },
                "required": ["timezone"]
            }
        }
    }
]

# Tool execution functions (replace with real API calls in production)
def get_current_weather(location: str, unit: str = "celsius") -> str:
    weather_data = {"Beijing": "Sunny 25°C", "Shanghai": "Cloudy 22°C", "Shenzhen": "Rainy 28°C"}
    return weather_data.get(location, f"Weather unknown for {location}")

def get_time(timezone: str) -> str:
    from datetime import datetime
    return datetime.now().strftime(f"%Y-%m-%d %H:%M:%S ({timezone})")

TOOL_MAP = {
    "get_current_weather": lambda **kw: get_current_weather(**kw),
    "get_time": lambda **kw: get_time(**kw)
}

def run_turn(messages, turn_num):
    """Execute a single user turn: call model, run tools in a loop until final answer."""
    request_num = 0
    while True:
        request_num += 1
        print(f"\nRequest {turn_num}-{request_num}:")

        response = client.chat.completions.create(
            model="mimo-v2.6-pro",
            messages=messages,
            tools=tools,
            extra_body={"thinking": {"type": "enabled"}}
        )

        assistant_message = response.choices[0].message
        messages.append(assistant_message)

        # Print full model response
        print(f"reasoning_content: {assistant_message.reasoning_content}")
        print(f"content: \"{assistant_message.content}\"")
        print(f"tool_calls: {assistant_message.tool_calls}")

        # If no tool calls, we have the final answer
        if not assistant_message.tool_calls:
            break

        # Execute each tool call and append results
        for tool_call in assistant_message.tool_calls:
            func_name = tool_call.function.name
            func_args = json.loads(tool_call.function.arguments)
            result = TOOL_MAP[func_name](**func_args)

            print(f"-> Tool result [{func_name}]: {result}")
            messages.append({
                "role": "tool",
                "tool_call_id": tool_call.id,
                "content": result
            })

# --- Multi-turn conversation ---
messages = []

# Turn 1
print("=== Turn 1 ===")
messages.append({"role": "user", "content": "How is the weather in Beijing today? What time is it now?"})
run_turn(messages, turn_num=1)

# Turn 2: reasoning_content from Turn 1 is already in messages via assistant_message
print("\n=== Turn 2 ===")
messages.append({"role": "user", "content": "How about Shanghai? And is it hotter or colder than Beijing?"})
run_turn(messages, turn_num=2)
```

{/* feishu-style:text-align:left */}
**Example Output**

{/* feishu-style:text-align:left */}
**Turn 1**: The user asks about the weather in Beijing and the current time. After receiving the user message, the model thinks and decides to call both `get_current_weather` and `get_time` tools simultaneously (Request 1-1). The client executes the tools and appends the results as `role: "tool"` messages to `messages`, then requests the model again. The model generates the final answer based on the tool results (Request 1-2).

```bash
=== Turn 1 ===

Request 1-1:
reasoning_content: The user wants both weather and time in Beijing. I can call both in parallel. Weather: get_current_weather(location: Beijing). Time: Beijing timezone is Asia/Shanghai. Call both together.
content: ""
tool_calls: [ChatCompletionMessageFunctionToolCall(id='call_01e402113df94ebfb85e3056', function=Function(arguments='{"location": "Beijing"}', name='get_current_weather'), type='function'), ChatCompletionMessageFunctionToolCall(id='call_f2bf1ab9f73842baa64e93d3', function=Function(arguments='{"timezone": "Asia/Shanghai"}', name='get_time'), type='function')]
-> Tool result [get_current_weather]: Sunny 25°C
-> Tool result [get_time]: 2026-09-22 00:38:45 (Asia/Shanghai)

Request 1-2:
reasoning_content: Present both results clearly.
content: "Here's the current information for Beijing:

- **Weather:** Sunny, 25°C (77°F) ☀️
- **Current time:** 00:38 AM, Tuesday, September 22, 2026 (Asia/Shanghai time)

It's a warm, clear day in Beijing — though it's just past midnight there right now, so the 25°C reading is more reflective of the warm night. Is there anything else you'd like to know?"
tool_calls: None
```

{/* feishu-style:text-align:left */}
**Turn 2**: The user follows up asking about Shanghai's weather and comparing it with Beijing. Since the first turn's `assistant` message (containing `reasoning_content`, `content`, `tool_calls`) has been accumulated into the conversation history via `messages.append()`, the model can directly access Beijing's 25°C information from the context, and only needs to call `get_current_weather` for Shanghai (Request 2-1), then provides a comparison answer based on both cities' weather data (Request 2-2).

```bash
=== Turn 2 ===

Request 2-1:
reasoning_content: Need weather and time for Shanghai, and compare to Beijing's 25°C. Call weather and time for Shanghai in parallel. The comparison only needs weather, but the previous pattern included time — the user asked "How about Shanghai?" which could mean both. I'll fetch both to be thorough, or just weather since the question focuses on comparison. I'll get both since it's ambiguous and follows the previous context.
content: ""
tool_calls: [ChatCompletionMessageFunctionToolCall(id='call_7e1e7118be8a49ef9de2e8f8', function=Function(arguments='{"location": "Shanghai"}', name='get_current_weather'), type='function'), ChatCompletionMessageFunctionToolCall(id='call_3ea278a5658443258c92b3de', function=Function(arguments='{"timezone": "Asia/Shanghai"}', name='get_time'), type='function')]
-> Tool result [get_current_weather]: Cloudy 22°C
-> Tool result [get_time]: 2026-09-22 00:38:51 (Asia/Shanghai)

Request 2-2:
reasoning_content: Shanghai: cloudy, 22°C vs Beijing sunny 25°C. Shanghai is 3°C colder. Both cities share the same timezone (Asia/Shanghai), so the time is the same.
content: "Here's the current information for Shanghai:

- **Weather:** Cloudy, 22°C (72°F) ☁️
- **Current time:** 00:38 AM, Tuesday, September 22, 2026 (Note: Shanghai and Beijing are in the same timezone, so the time is identical in both cities)

**Comparison:** Shanghai is **colder** than Beijing right now — 22°C vs. 25°C, a difference of about 3°C (5°F). Beijing is also sunny while Shanghai is cloudy. Anything else you'd like to check?"
tool_calls: None
```


--- DOCUMENT: Structured Output ---
URL: https://mimo.mi.com/static/docs/quick-start/usage-guide/text-generation/structured-output.md

# Structured Outputs

{/* feishu-style:text-align:left */}
Structured Output (JSON mode) enables models to generate responses in a specified JSON format, ensuring outputs are controllable and easy to parse. It is suitable for scenarios requiring structured data, such as data extraction, form filling, classification and tagging, and API response formatting.

## Supported Models

{/* feishu-style:text-align:left */}
Currently, the `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro` and `mimo-v2.5` models are supported.

## Request Parameters

- `response_format`: Response format control parameter. Pass `{"type": "json_object"}` to enable JSON mode output.

- `messages`: You must explicitly instruct the model in the system or user message to return only JSON, and fully define the fields, hierarchy, and data types of the expected JSON structure. Providing examples is recommended.

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

Plan the value of `max_completion_tokens` carefully. Setting it too low may cause the output to be truncated, resulting in incomplete JSON that cannot be parsed.

</div>

## Call Examples

### Basic Call

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
import json
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.\nDo not add any explanations outside JSON. Parse meeting information and return nested structured JSON: {\"meeting_meta\": {\"meeting_topic\": string | null, \"start_time\": string | null, \"participants\": array<string>}, \"action_items\": [{\"task\": string, \"responsible_person\": string, \"deadline\": string}]}\nFill null for unknown fields."
        },
        {
            "role": "user",
            "content": "Product iteration meeting at 14:00 tomorrow. Attendees: Li Ming, Wang Hua. Task: Complete API document revision by Friday, taken by Wang Hua."
        }
    ],
    response_format={
        "type": "json_object"
    }
)

result = completion.choices[0].message.content
try:
    parsed_json_data = json.loads(result)
    print(json.dumps(parsed_json_data, indent=4, ensure_ascii=False))
except json.JSONDecodeError as e:
    print(f"JSON parsing failed: {str(e)}")
    print(f"Raw complete content: {result}")
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.\nDo not add any explanations outside JSON. Parse meeting information and return nested structured JSON: {\"meeting_meta\": {\"meeting_topic\": string | null, \"start_time\": string | null, \"participants\": array<string>}, \"action_items\": [{\"task\": string, \"responsible_person\": string, \"deadline\": string}]}\nFill null for unknown fields."
        },
        {
            "role": "user",
            "content": "Product iteration meeting at 14:00 tomorrow. Attendees: Li Ming, Wang Hua. Task: Complete API document revision by Friday, taken by Wang Hua."
        }
    ],
    "response_format": {
        "type": "json_object"
    }
}'
```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Response Example**

```json
{
    "id": "d7bbc231-ed6a-4884-a9e4-d7caff96cf25_c37a9ce97e9542079b9d948b07c10209",
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "message": {
                "content": "{\"meeting_meta\": {\"meeting_topic\": \"Product iteration meeting\", \"start_time\": \"2025-12-17 14:00\", \"participants\": [\"Li Ming\", \"Wang Hua\"]}, \"action_items\": [{\"task\": \"Complete API document revision\", \"responsible_person\": \"Wang Hua\", \"deadline\": \"2025-12-19\"}]}",
                "role": "assistant",
                "tool_calls": null,
                "reasoning_content": "The user wants me to parse meeting information and return structured JSON.\n\nMeeting: Product iteration meeting\nTime: 14:00 tomorrow\nParticipants: Li Ming, Wang Hua\nTask: Complete API document revision by Friday, responsible: Wang Hua\n\nI need to fill unknown fields with null. Today's date is Tuesday, December 16, 2025, so tomorrow is December 17, 2025. But I should probably just put what was given: \"14:00 tomorrow\" — I could interpret it. The instruction says to fill null for unknown fields. \"Tomorrow\" relative to today (2025-12-16) is 2025-12-17. I'll provide \"2025-12-17 14:00\" as start_time. Actually, the user message just says \"at 14:00 tomorrow\" — it's reasonable to resolve it. Friday relative to 2025-12-17 would be 2025-12-19. I'll provide that.\n\nI'll return JSON only, no explanations outside JSON."
            }
        }
    ],
    "created": 1790008400,
    "model": "mimo-v2.6-pro",
    "object": "chat.completion",
    "usage": {
        "completion_tokens": 344,
        "prompt_tokens": 161,
        "total_tokens": 505,
        "completion_tokens_details": {
            "reasoning_tokens": 253
        },
        "prompt_tokens_details": {
            "cached_tokens": 0
        }
    }
}
```

### Streaming Output

<Tab>
  <TabItem label={`Python SDK`}>

> In streaming mode, the model outputs JSON content incrementally. You must concatenate the complete JSON string on the client side before parsing to avoid failures caused by truncation.

```python
import os
import json
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.\nReturn only compact JSON without any extra explanations. Use the specified format: {\"destination\": string, \"trip_days\": number, \"core_activity\": array<string>, \"budget_level\": \"low\"|\"mid\"|\"high\"}"
        },
        {
            "role": "user",
            "content": "Plan a 3-day city trip focusing on museums and local food with a moderate budget."
        }
    ],
    response_format={
        "type": "json_object"
    },
    stream=True
)

full_json_content = ""
for chunk in completion:
    if not chunk.choices:
        continue
    message_delta = chunk.choices[0].delta
    if message_delta.content:
        full_json_content += message_delta.content

try:
    parsed_json_data = json.loads(full_json_content)
    print(json.dumps(parsed_json_data, indent=4, ensure_ascii=False))
except json.JSONDecodeError as e:
    print(f"JSON parsing failed: {str(e)}")
    print(f"Raw complete content: {full_json_content}")

```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.\nReturn only compact JSON without any extra explanations. Use the specified format: {\"destination\": string, \"trip_days\": number, \"core_activity\": array<string>, \"budget_level\": \"low\"|\"mid\"|\"high\"}"
        },
        {
            "role": "user",
            "content": "Plan a 3-day city trip focusing on museums and local food with a moderate budget."
        }
    ],
    "response_format": {
        "type": "json_object"
    },
    "stream": true
}'

```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Response Example (Concatenated Streaming Result)** 

```json
{
    "destination": "Lisbon, Portugal",
    "trip_days": 3,
    "core_activity": [
        "Visit the National Museum of Ancient Art",
        "Explore the MAAT and Belém cultural district",
        "Wander the Calouste Gulbenkian Museum",
        "Taste pastéis de nata at Pastéis de Belém",
        "Sample petiscos and seafood at Time Out Market",
        "Try bacalhau and bifana in local tascas",
        "Walk through Alfama with fado dinner"
    ],
    "budget_level": "mid"
}
```

## How to Get Accurate JSON Output

{/* feishu-style:text-align:left */}
The `json_object` mode only guarantees syntactically valid JSON output. The actual data structure is entirely defined by your prompt. The clearer and more complete your prompt constraints are, the more closely the model's JSON output will match your expectations.

{/* feishu-style:text-align:left */}
**1. Enforce JSON-Only Output**

{/* feishu-style:text-align:left */}
Explicitly require the model to return JSON only, with no explanations, comments, or Markdown code blocks, to prevent parsing errors caused by extra text.

{/* feishu-style:text-align:left */}
**2. Provide a Complete JSON Structure Template**

{/* feishu-style:text-align:left */}
List all fields, data types, and nesting levels completely. Adding examples makes it easier to constrain the model:

```json
{
  "name": string,
  "count": number,
  "tags": string[],
  "date": "YYYY-MM-DD"
}
```

{/* feishu-style:text-align:left */}
**3. Constrain Enum Values and Numeric Ranges**

- Use enum annotations for fixed options: `"status": "active" | "inactive" | "pending"`

- Constrain numeric ranges: `score: number (0-100)`

{/* feishu-style:text-align:left */}
**4. Define Null Value Handling Rules**

{/* feishu-style:text-align:left */}
Specify in advance that unknown fields should be filled with `null`, e.g., in your prompt: fill null for unknown fields.

{/* feishu-style:text-align:left */}
**5. Ensure Reliable JSON Output**

{/* feishu-style:text-align:left */}
Enabling `json_object` mode only guarantees syntactically valid JSON; it cannot enforce that returned fields, data types, and nesting match your predefined structure. In production environments, it is recommended to use the `jsonschema` library for strict structural validation of model output. If validation fails, you can retry with an enhanced prompt, apply business-level fallback logic, or take other contingency measures.

{/* feishu-style:text-align:left */}
Below is a Python example. The scenario: parse customer service conversation text, automatically extract core ticket information, and implement automatic ticket classification, urgency level assignment, and intelligent dispatch.

```python
import json
import os
from jsonschema import validate, ValidationError
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

ticket_schema = {
    "type": "object",
    "properties": {
        "problem_type": {
            "type": "string",
            "enum": ["device_fault", "consult", "complaint", "other"]
        },
        "device_model": {"type": ["string", "null"]},
        "urgent_level": {
            "type": "string",
            "enum": ["low", "normal", "high"]
        },
        "user_requirements": {"type": "array", "items": {"type": "string"}}
    },
    "required": ["problem_type", "device_model", "urgent_level", "user_requirements"]
}

def extract_ticket(text: str):
    sys_prompt = (
        "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.\n"
        "Return JSON only, no explanations, no extra text.\n"
        "Format: {\"problem_type\":\"device_fault|consult|complaint|other\",\"device_model\":string|null,\"urgent_level\":\"low|normal|high\",\"user_requirements\":array<string>}\n"
        "Fill null if device model is unknown."
    )
    messages = [
        {"role": "system", "content": sys_prompt},
        {"role": "user", "content": text}
    ]

    resp = client.chat.completions.create(
        model="mimo-v2.6-pro",
        messages=messages,
        response_format={"type": "json_object"},
        stream=False
    )
    json_text = resp.choices[0].message.content
    data = json.loads(json_text)

    try:
        validate(instance=data, schema=ticket_schema)
        return {"ok": True, "data": data}
    except ValidationError as e:
        return {"ok": False, "error": e.message, "raw": data}

if __name__ == "__main__":
    # Test with two sample tickets
    input_list = [
        "My speaker can't connect Wi-Fi, I restarted it but no use, hope to solve quickly, don't know the model.",
        "Want to ask how to set timing function, device model is Watch S1, no urgent demand."
    ]

    for idx, text in enumerate(input_list, 1):
        print(f"==== Ticket {idx} extraction result ====")
        res = extract_ticket(text)
        if res["ok"]:
            print(json.dumps(res["data"], indent=4))
        else:
            print(f"Validation failed: {res['error']}")

```


--- DOCUMENT: Batch Inference (Batch API) ---
URL: https://mimo.mi.com/static/docs/quick-start/usage-guide/text-generation/batch-api.md

# Batch API

{/* feishu-style:text-align:left */}
Batch Inference (Batch API) is an offline, large-scale data processing solution for workloads that do not require real-time responses. The interface is OpenAI-compatible and is well suited for scenarios such as model evaluation, data labeling, and batch regression testing. You can create asynchronous jobs via the SDK or the console. Key advantages:

- ⭐ **Lower cost**: Batch Inference (Batch API) is priced at 50% of the real-time API

> Models supported for batch inference: `mimo-v2.6-pro`, `mimo-v2.6-flash`,when using the batch inference service for model calls, the model name must be in lowercase.

- ⭐ **Off-peak smart scheduling**: After a job is submitted, the system automatically schedules it for execution during off-peak hours, making full use of idle compute

- ⭐ **Flexible usage**: Two ways to use it — API and console — supporting full lifecycle job management through either the API or the Open Platform console

- ⭐ **OpenAI compatible**: The API protocol is identical to OpenAI's, making migration from OpenAI Batch simple

## Applicable Scenarios

<table>
<colgroup>
<col style="width: 28.05%" />
<col style="width: 71.95%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Scenario</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Description</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Data labeling**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Label large volumes of text and images</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Model evaluation**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Benchmark testing, regression validation</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Content moderation**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Offline batch classification and filtering</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Batch generation**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Summarization, translation, structured extraction</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Academic research**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Large-scale data experiments, research data processing</span></td>
</tr>
</tbody>
</table>

## Batch Inference API Pricing

{/* feishu-style:text-align:left */}
Xiaomi MiMo API pay-as-you-go billing uses a regular Open Platform API Key and deducts from your account balance based on actual token usage. It is not interchangeable with Token Plan package quotas.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Billing notes

- Batch Inference (Batch API) price = real-time API price × 50%

- Billing unit: CNY per million tokens (Mainland China); USD per million tokens (Overseas)

- Cache hits: When the request's prefix content hits the Prompt Cache, billing uses the cache-hit price

- Batch inference currently supports the `mimo-v2.6-pro` and `mimo-v2.6-flash` models. Pricing for both models in Mainland China and overseas is listed below

- API Key format for batch inference: sk-xxxxx. Go to [API Keys](https://platform.xiaomimimo.com/console/api-keys) to create an API Key.

</div>

{/* feishu-style:text-align:left */}
**Mainland China pricing**

<table>
<colgroup>
<col style="width: 21%" />
<col style="width: 13.28%" />
<col style="width: 13.28%" />
<col style="width: 13.41%" />
<col style="width: 13.41%" />
<col style="width: 13.41%" />
<col style="width: 12.210000000000008%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Inference type</span></th>
<th colspan="3"><span style="display: inline-block; width: 100%; text-align: left;">Real-time Inference API</span></th>
<th colspan="3"><span style="display: inline-block; width: 100%; text-align: left;">Batch Inference API</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">MiMo-V2.6 Series</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Input (cache hit)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Input (cache miss)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Output</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Input (cache hit)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Input (cache miss)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Output</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.025</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥3.00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥6.00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.0125</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥1.50</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥3.00</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.02</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥1.00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥2.00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.01</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.50</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥1.00</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
**Overseas pricing**

<table>
<colgroup>
<col style="width: 21%" />
<col style="width: 13.28%" />
<col style="width: 13.41%" />
<col style="width: 13.41%" />
<col style="width: 13.41%" />
<col style="width: 13.41%" />
<col style="width: 12.080000000000013%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Inference type</span></th>
<th colspan="3"><span style="display: inline-block; width: 100%; text-align: left;">Real-time Inference API</span></th>
<th colspan="3"><span style="display: inline-block; width: 100%; text-align: left;">Batch Inference API</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">MiMo-V2.6 Series</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Input (cache hit)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Input (cache miss)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Output</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Input (cache hit)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Input (cache miss)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Output</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.0036</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.435</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.87</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.0018</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.2175</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.435</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.0028</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.14</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.28</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.0014</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.07</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.14</span></td>
</tr>
</tbody>
</table>

## Getting Started with Batch Inference

### Prerequisites

{/* feishu-style:text-align:left */}
Before using Batch Inference (Batch API), complete the following steps:

<table>
<colgroup>
<col style="width: 32.32%" />
<col style="width: 67.68%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Step</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Description</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">1. Register an account</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Register a Xiaomi MiMo Open Platform account</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">2. Real-name verification</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Complete real-name verification</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">3. Top up balance</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Top up your account balance (Batch API deducts from your balance)</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">4. Get an API Key</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Create a usable API Key</span><br /><blockquote>- Batch task Created via API: use the API Key you created<br />- Batch task Created via the console: attached to the platform's default API Key by default; no need to use your own API Key</blockquote></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">5. Get the Base URL</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Get the Base URL,Go to the [Batch Inference](https://platform.xiaomimimo.com/console/batch) page to get it</span></td>
</tr>
</tbody>
</table>

### Step 1: Prepare the File

{/* feishu-style:text-align:left */}
Prepare your data file in JSONL format (JSON Lines — one JSON object per line) following the file format requirements below. Each line contains the details of a single API request.

#### File Examples

{/* feishu-style:text-align:left */}
Below is an example of an input file containing 2 requests (a `.jsonl` file), corresponding to the 3 supported endpoint formats. **Note: when using batch inference via the console, only the OpenAI | completions file format is currently supported for upload. Using the API is more flexible and supports all 3 file formats.** 

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Base URL: Go to the [Batch Inference](https://platform.xiaomimimo.com/console/batch) page to get it

</div>

{/* feishu-style:text-align:left */}
**OpenAI | completions,** [**download sample file**](https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/others/input-completions.jsonl)

```json
{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "mimo-v2.6-flash", "messages": [{"role": "user", "content": "Hello"}]}}
{"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "mimo-v2.6-pro", "messages": [{"role": "user", "content": "World"}]}}
```

{/* feishu-style:text-align:left */}
**OpenAI | responses,** [**download sample file**](https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/others/input-responses-update.jsonl)

```json
{"custom_id": "request-1", "method": "POST", "url": "/v1/responses", "body": {"model": "mimo-v2.6-pro", "input": "Hello"}}
{"custom_id": "request-2", "method": "POST", "url": "/v1/responses", "body": {"model": "mimo-v2.6-flash", "input": "please introduce yourself"}}
```

{/* feishu-style:text-align:left */}
**Anthropic | messages,** [**download sample file**](https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/others/input-claude-update.jsonl)

```json
{"custom_id": "request-1", "method": "POST", "url": "/anthropic/v1/messages", "body": {"model": "mimo-v2.6-pro", "messages": [{"role": "user", "content": [{"type": "text", "text": "Hello"}]}]}}
{"custom_id": "request-2", "method": "POST", "url": "/anthropic/v1/messages", "body": {"model": "mimo-v2.6-flash", "messages": [{"role": "user", "content": [{"type": "text", "text": "please introduce yourself"}]}]}}
```

#### File Specification Requirements

- **Basic file requirements**

{/* feishu-style:text-align:left */}
Default maximum file size is 128 MB. A single file may only contain requests for one batch inference endpoint.

- **Field rules**

   - Every request must include a `custom_id` field, **a string that is unique within the file**, used to map each request to its corresponding result.

   - Each request is sent and returns its result independently. If multiple requests share the same prompt, include the same prompt in each request.

   - The `body` field of each request must be consistent with the request body of the underlying model invocation API and must be a valid JSON Object.

### Step 2: Create a Batch Inference Job

1. On the [Batch Inference](https://platform.xiaomimimo.com/console/batch) page , click **Create Batch Inference Job**.

1. On the create job page, upload your JSONL file, fill in a **job description**, and set the **maximum wait time** (1–14 days).

> **When using batch inference via the console, only the OpenAI | completions file format is currently supported for upload. Using the API is more flexible and supports all 3 file formats.** 
>
> - OpenAI | completions
>
> - OpenAI | responses
>
> - Anthropic | messages

1. When finished, click **Create**.

### Step 3: Manage Jobs

1. **View**:

   - On the job list page, view the **progress** (processed requests / total requests) and **status** of each job. Or enter the job detail page for more information.

   - Search by job description or ID to quickly locate a target job.

1. **Manage**:

   - Cancel: Jobs in the "In Progress" state can be cancelled from the **Actions** column.

   - Troubleshoot errors: "Failed" jobs allow you to download the error file for details. File downloads are available from both the job list Actions column and the job detail page.

### Step 4: Download Results

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

The system retains your data for only 30 days. Please download and back up your data promptly. After expiry, files are automatically deleted and cannot be recovered.

</div>

{/* feishu-style:text-align:left */}
After a job completes, files can be downloaded from the job list Actions column and the job detail page:

- **Success file**: records all successful requests and their `response` results.

- **Error file (if any)** : records all failed requests and their `error` details.

{/* feishu-style:text-align:left */}
Both files include the `custom_id` field, used to match against your original input data, associate results, or locate errors.

### Step 5: View Usage Statistics (Optional)

{/* feishu-style:text-align:left */}
On the [Billing Details](https://platform.xiaomimimo.com/console/usage) page, filter and view usage statistics for batch inference.

{/* feishu-style:text-align:left */}
**View data overview**: **Select a time range**, set **Inference Type** to **Batch Inference**, select an **API Key**, and view the batch inference model invocation overview.

> - Jobs created via the API: use the API Key you created; select that Key to view usage
>
> - Jobs created via the console: attached to the platform's default API Key by default; no need to use your own API Key; select **Other** to view usage

## API Reference

### Step 1: Upload the Job File to the File Service

{/* feishu-style:text-align:left */}
You can use Curl to upload the job file to the file service's bucket. The gateway platform will subsequently read the request information from the file for batch inference.

{/* feishu-style:text-align:left */}
**Request example**

{/* feishu-style:text-align:left */}
API Key format for batch inference: sk-xxxxx. Go to [API Keys](https://platform.xiaomimimo.com/console/api-keys) to create an API Key.

```json
curl https://batch-api-${region}.xiaomimimo.com/v1/files \
-H "Authorization: Bearer $ARK_API_KEY" \
-F 'purpose=batch' \ 
-F 'file=@/Users/doc/demo.jsonl' \     # file path
```

{/* feishu-style:text-align:left */}
**Request parameters**

- purpose — Upload file category

   - batch (batch inference file)

- file=@ — Local path of the file to upload

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

The default file expiration time is currently 30 days

</div>

{/* feishu-style:text-align:left */}
**Sample request response**

```json
{
    "id": "file-8a761b15a195",
    "object": "file",
    "purpose": "batch",
    "filename": "input.jsonl",
    "bytes": 327,
    "status": "active",
    "error": null,
    "metadata": null,
    "mime_type": "application/jsonl",
    "created_at": 1780034477,
    "expire_at": 1780038877,
    "preprocess_configs": null
}
```

<table>
<colgroup>
<col style="width: 19.91%" />
<col style="width: 80.09%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Field</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Description</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">id</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Unique file identifier, formatted as `file-` + UUID prefix</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">object</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Object type; fixed value `"file"`, indicating this is a file resource</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">purpose</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">File category. `"batch"` indicates use for batch processing</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">filename</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Original file name</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">bytes</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">File size in **bytes**</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">status</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">File status. `"active"` means the file is available; other possible values include `"pending"` (processing) and `"error"` (failed)</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">error</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Error message; null when there is no error; contains the specific error description on failure</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">metadata</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Custom metadata — user-supplied key-value pairs; not currently used</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mime_type</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">MIME type</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">created_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Creation time, Unix timestamp (seconds)</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">expire_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Expiration time, Unix timestamp. The file may be automatically cleaned up after expiry</span></td>
</tr>
</tbody>
</table>

### Step 2: Create a Batch Inference Job

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Custom job timeout is supported, ranging from 1 to 14 days

</div>

{/* feishu-style:text-align:left */}
**Request example**

```bash
curl -X POST https://batch-api-${region}.xiaomimimo.com/v1/batches \
  -H "Authorization: Bearer $ARK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input_file_id": "$FILE_ID",
    "endpoint": "/v1/chat/completions",
    "completion_window": "24h",
    "name": "test"
  }'
```

{/* feishu-style:text-align:left */}
**Response example**

```json
{
  "id": "batch_c089f12dde664f07aeadd8c7",
  "object": "batch",
  "endpoint": "/v1/chat/completions",
  "errors": null,
  "input_file_id": "file-0f4add6a5001",
  "completion_window": "24h",
  "status": "validating",
  "output_file_id": null,
  "error_file_id": null,
  "created_at": 1711402400,
  "in_progress_at": null,
  "expires_at": 1711488800,
  "finalizing_at": null,
  "completed_at": null,
  "failed_at": null,
  "expired_at": null,
  "cancelling_at": null,
  "cancelled_at": null,
  "request_counts": { "total": 0, "completed": 0, "failed": 0 }
}
```

<table>
<colgroup>
<col style="width: 35.89%" />
<col style="width: 18.3%" />
<col style="width: 45.81%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Field**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Type**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Description**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">id</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">string</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Unique Batch Job identifier, formatted as `batch_` + random ID</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">object</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">string</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Object type, fixed as `"batch"`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">endpoint</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">string</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">API endpoint for this batch call, e.g. `/v1/chat/completions`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">errors</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">object/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Error message; null when there is no error</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">input_file_id</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">string</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Input file ID, pointing to the uploaded `.jsonl` file</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">completion_window</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">string</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Completion window; currently fixed at `"24h"`. Becomes `expired` on timeout</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">status</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">string</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Current status (9 possible values — see the previous message)</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">output_file_id</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">string/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Output file ID; populated only after completion; can be used to download results</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">error_file_id</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">string/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Error file ID; populated only when there are failed requests; records the reason for each failure</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">created_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Creation time (Unix timestamp, seconds)</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">in_progress_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Time the job entered processing; null when processing has not started</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">expires_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Expiration time, i.e. `created_at + completion_window`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">finalizing_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Time the job entered the finalizing stage</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">completed_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Completion time; populated only in the `completed` state</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">failed_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Failure time; populated only in the `failed` state</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">expired_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Expiration time; populated only in the `expired` state</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">cancelling_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Time the cancellation was initiated</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">cancelled_at</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int/null</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Time the cancellation completed; populated only in the `cancelled` state</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">request_counts</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">object</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Request counter</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">request_counts.total</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Total number of requests</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">request_counts.completed</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Number of completed requests</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">request_counts.failed</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">int</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Number of failed requests</span></td>
</tr>
</tbody>
</table>

### Step 3: Query Batch Inference Job Status

{/* feishu-style:text-align:left */}
**Request example**

```bash
curl https://batch-api-${region}.xiaomimimo.com/v1/batches/$BATCH_ID \
  -H "Authorization: Bearer $ARK_API_KEY"
```

{/* feishu-style:text-align:left */}
Batch inference job statuses and their descriptions:

<table>
<colgroup>
<col style="width: 26.56%" />
<col style="width: 23.67%" />
<col style="width: 49.769999999999996%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Status</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Status code</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Description</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Initializing</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">validating</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">The job is initializing.</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Running</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">in_progress</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">The job is running.</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Completed</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">completed</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">The job has fully completed.</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Failed</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">failed</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Job execution failed, possibly due to timeout or other reasons.</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Cancelling</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">cancelling</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">The user is actively cancelling the job</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Cancelled</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">cancelled</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">The user's cancellation succeeded; the job has been terminated</span></td>
</tr>
</tbody>
</table>

### Step 4: Download Batch Inference Job Results

#### Success file (output_file)

{/* feishu-style:text-align:left */}
One result object per line, mapped to the input via `custom_id`:

```bash
{"id":"batch_req_xxx","custom_id":"request-1","response":{"status_code":200,"body":{"id":"chatcmpl-xxx","choices":[{"message":{"content":"Hello!"}}]}},"error":null}
{"id":"batch_req_yyy","custom_id":"request-2","response":{"status_code":200,"body":{"id":"chatcmpl-yyy","choices":[{"message":{"content":"World!"}}]}},"error":null}
```

#### Error file (error_file)

{/* feishu-style:text-align:left */}
Generated only when there are failed requests:

```bash
{"id":"batch_req_zzz","custom_id":"request-3","response":null,"error":{"code":"inference_failed","message":"400 Bad Request"}}
```

{/* feishu-style:text-align:left */}
**The input** `**custom_id**` **appears verbatim in the output**, used to align inputs with outputs.

{/* feishu-style:text-align:left */}
After the batch inference job finishes running, you can download the result files via curl. The result files fall into 2 categories:

{/* feishu-style:text-align:left */}
**Download the output file**

```bash
curl https://batch-api-${region}.xiaomimimo.com/v1/files/${result_file_id}/content \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -o output.jsonl
```

{/* feishu-style:text-align:left */}
Output file format (JSONL, one result per line):

```json
{"id":"f7ed2","custom_id":"request-1","response":{"status_code":200,"body":{"id":"chatcmpl-xxx","choices":[{"message":{"content":"Hello!"}}]}},"error":null}
{"id":"a3b21","custom_id":"request-2","response":{"status_code":200,"body":{"id":"chatcmpl-yyy","choices":[{"message":{"content":"World!"}}]}},"error":null}
```

{/* feishu-style:text-align:left */}
**Download the error file**

{/* feishu-style:text-align:left */}
If there are failed requests, `error_file_id` is not empty:

```bash
curl https://batch-api-${region}.xiaomimimo.com/v1/files/${error_file_id}/content \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -o errors.jsonl
```

{/* feishu-style:text-align:left */}
Error file format:

```json
{"id":"db7da","custom_id":"request-3","response":null,"error":{"code":"inference_failed","message":"400 Bad Request"}}
```

### Other API Examples

#### Cancel a Batch Inference Job

{/* feishu-style:text-align:left */}
**Request example**

```bash
curl -X POST https://batch-api-${region}.xiaomimimo.com/v1/batches/batch_c089f12dde664f07aeadd8c7/cancel \
  -H "Authorization: Bearer YOUR_API_KEY"
```

{/* feishu-style:text-align:left */}
**Response example**

```json
{
  "id": "batch_c089f12dde664f07aeadd8c7",
  "object": "batch",
  "endpoint": "/v1/chat/completions",
  "errors": null,
  "input_file_id": "file-0f4add6a5001",
  "completion_window": "24h",
  "status": "cancelling",
  "output_file_id": null,
  "error_file_id": null,
  "created_at": 1711402400,
  "in_progress_at": 1711402410,
  "expires_at": 1711488800,
  "finalizing_at": null,
  "completed_at": null,
  "failed_at": null,
  "expired_at": null,
  "cancelling_at": 1711402600,
  "cancelled_at": null,
  "request_counts": { "total": 2, "completed": 1, "failed": 0 }
}
```

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Cancellation is an asynchronous operation; the returned status is `cancelling`. On subsequent queries the status becomes `cancelled`, at which point `output_file_id` and `error_file_id` are not null (the portion already completed generates an output file).

</div>

#### OpenAI SDK Compatibility

{/* feishu-style:text-align:left */}
Batch Inference (Batch API) is OpenAI-protocol compatible and can be used directly with the OpenAI Python SDK:

```python
from openai import OpenAI

client = OpenAI(
    api_key="your-mimo-api-key",
    base_url="https://batch-api-${region}.xiaomimimo.com/v1"
)

# Upload the file
file = client.files.create(
    file=open("input.jsonl", "rb"),
    purpose="batch"
)

# Create the batch
batch = client.batches.create(
    input_file_id=file.id,
    endpoint="/v1/chat/completions",
    completion_window="24h"
)

# Query status
batch = client.batches.retrieve(batch.id)
print(f"Status: {batch.status}")
print(f"Completed: {batch.request_counts.completed}/{batch.request_counts.total}")

# Download results
if batch.output_file_id:
    result = client.files.content(batch.output_file_id)
    with open("output.jsonl", "wb") as f:
        f.write(result.content)
```

## FAQ

### Why doesn't the job start immediately after submission?

{/* feishu-style:text-align:left */}
Batch Inference (Batch API) uses an **off-peak scheduling** strategy: the system schedules jobs automatically based on online resource availability, with no manual intervention required. When resources are tight, job startup and execution may be delayed.

### How do I retry a failed job?

{/* feishu-style:text-align:left */}
Batch Inference (Batch API) currently does **not support in-job retry or resume**. If a job fails, you need to:

1. Download the error file;

1. Review the failure reasons in the error file;

1. Create a new job and resubmit.

### What should I do if the job is partially successful?

1. If some requests in your job succeeded and others failed, the **success file** contains all successful request results, and the **error file** contains the failed requests and their error reasons. Successfully completed requests are billed normally.

1. You can download the error file, review the failure reasons, and create a new job to resubmit.

### What happens if my balance is insufficient?

1. Balance is 0 at job creation: you can still browse and go through the job creation flow; the job creation will fail

1. Balance becomes insufficient while the job is running: the job fails. The portion already completed generates a success file and is billed normally; the portion not executed generates a failure file and is not charged

### Can I use it without real-name verification?

{/* feishu-style:text-align:left */}
**No.**  Real-name verification must be completed before using Batch Inference (Batch API). Unverified users who visit the batch inference page will see a guide page and be redirected to the real-name verification page.

### Is streaming output supported?

{/* feishu-style:text-align:left */}
**No.**  Batch scenarios are asynchronous processing and do not apply to the streaming protocol.

### Is Token Plan deduction supported?

{/* feishu-style:text-align:left */}
**No.**  Batch Inference (Batch API) only deducts from your account's cash balance based on actual token usage and is not interchangeable with Token Plan package quotas.

### How long are result files retained?

{/* feishu-style:text-align:left */}
Input files and result files are retained for **30 days by default** and are automatically cleaned up after expiry. Please download them promptly.

### How are completed, failed, and cancelled jobs billed?

- Only successful requests are billed. Requests that fail during file parsing or job execution **incur no charges**

- **Requests successfully completed before cancellation are billed normally**


--- DOCUMENT: Image Understanding ---
URL: https://mimo.mi.com/static/docs/quick-start/usage-guide/multimodal-understanding/image-understanding.md

# Image Understanding

{/* feishu-style:text-align:left */}
The image understanding model can answer based on the images you provide, supporting both image URL and Base64 encoding as input methods, and is suitable for scenarios such as image description and classification. 

## Quick Start

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

For preparations such as obtaining an API Key, please refer to [First API Call](https://platform.xiaomimimo.com/#/docs/quick-start/first-api-call).

</div>

{/* feishu-style:text-align:left */}
Quickly experience the effect of image understanding by passing the model through the image URL method. The sample code is as follows. 

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/image/image_example.png"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/image/image_example.png"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Response**

```json
{
    "id": "75a0045f-a28f-4b0f-8ddd-6c90c6ef6587_90e3b2e2dd1f410c8b5526a21526bdec",
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "message": {
                "content": "This is a serene, enchanted-looking ancient forest scene, rich with detail:\n1.  **Foreground:** At the bottom left, lush, large green ferns with delicate fronds grow on a bed of moss-covered smooth river stones. A shallow, clear stream flows gently over the stones, its surface glinting brightly as it catches golden sunlight. Mossy rocks and small, leafy green shrubs line the right bank, their leaves glowing in the warm light.\n2.  **Midground:** The stream winds deeper into the woods, surrounded by thick, vibrant moss and lush low greenery. Dominating the left is a massive, gnarled ancient tree with thick, exposed roots gripping the earth, its dark, twisted trunk curving outward. Other tall, mature trees with textured dark bark fill the space, creating a dense, old-growth forest feel.\n3.  **Light & Atmosphere:** Warm, golden sunbeams (crepuscular rays) filter down through the forest canopy, cutting through a soft mist that hangs in the air. Tiny glowing particles, like sun dust or fireflies, drift in the sunlit space, adding a magical, dreamlike quality. The distant forest fades into soft haze, giving the scene deep, peaceful atmosphere.\n4.  **Overall feel:** The image evokes a sense of quiet, untouched wilderness, peaceful and full of life, with the gentle flow of the stream and warm, dappled light creating an intimate, enchanted natural space.",
                "role": "assistant",
                "tool_calls": null,
                "reasoning_content": "Got it, let's break down this beautiful forest scene step by step.\n\nFirst, the overall mood: it's a misty, sun-dappled ancient forest, with a sense of quiet, magic (the little light specks feel like fireflies or sun particles), peaceful and lush.\n\nForeground: At the bottom left, large, vibrant green ferns with feathery fronds sit on moss-covered, smooth rounded stones. The stream here is shallow, with clear water flowing over the stones, catching the golden sunlight. Mossy rocks line the right side, with small leafy green bushes growing on them, their leaves catching the warm light.\n\nMidground: A gentle stream winds its way deeper into the forest, its surface glinting with the sunbeams. The banks are thick with moss, ferns, and low greenery, all lush and healthy. Large, ancient trees dominate here: on the left, a massive tree with thick, gnarled, exposed roots gripping the ground, its dark trunk curving outward. Other tall trees with dark, textured bark fill the space, their trunks disappearing into the mist.\n\nBackground & lighting: The scene is backlit by warm, golden sunbeams (crepuscular rays) that filter through the tree canopy, cutting through the mist that hangs in the air, creating a soft, hazy glow. The light catches tiny particles in the air, making them look like glowing embers or fireflies, adding a magical touch. The distant forest fades into soft mist, giving the scene depth.\n\nAll elements come together to feel like a serene, untouched, enchanted forest space, full of life and soft, warm light, with the sound of the gentle stream almost imaginable."
            }
        }
    ],
    "created": 1790007893,
    "model": "mimo-v2.6-pro",
    "object": "chat.completion",
    "usage": {
        "completion_tokens": 656,
        "prompt_tokens": 1085,
        "total_tokens": 1741,
        "completion_tokens_details": {
            "reasoning_tokens": 351
        },
        "prompt_tokens_details": {
            "cached_tokens": 1024,
            "image_tokens": 1024
        }
    }
}
```

## Supported models

{/* feishu-style:text-align:left */}
Currently, the `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed` and `mimo-v2.5` models are supported.

## Image Input Method

{/* feishu-style:text-align:left */}
Supported ways to upload images are as follows:

- Image URL Input: A publicly accessible image URL address must be provided. 

- Base64 Encoding Input: Convert the image to a Base64-encoded string before passing it in.

### Image URL Input

{/* feishu-style:text-align:left */}
Directly pass in the image via the publicly accessible image URL, which is suitable for scenarios where the image is already stored in a publicly accessible environment. The file size of a single image cannot exceed 50 MB.

#### OpenAI Chat Completions API

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/image/image_example.png"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/image/image_example.png"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

#### Anthropic Messages API

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/anthropic"
)

message = client.messages.create(
    model="mimo-v2.6-pro",
    max_tokens=1024,
    system="You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "source": {
                        "type": "url",
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/image/image_example.png"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ]
)

print(message.content)
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/anthropic/v1/messages' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "max_tokens": 1024,
    "system": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.",
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "source": {
                        "type": "url",
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/image/image_example.png"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ]
}'
```

  </TabItem>
</Tab>

### Base64 Encoded Input

{/* feishu-style:text-align:left */}
Convert the image file to a Base64-encoded string and then pass it in, which is suitable for scenarios where the image cannot be accessed via a public network URL. The size of the converted Base64-encoded string cannot exceed 50 MB.

> In the following example, both `{MIME_TYPE}` and `$BASE64_IMAGE` are placeholders, please replace them with actual valid data before running.

#### OpenAI Chat Completions API

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Please include the prefix before Base64 encoding: `data:{MIME_TYPE};base64,$BASE64_IMAGE`

- `{MIME_TYPE}`: The MIME type (media type) of the image, used to identify the image format, needs to be replaced with the MIME value corresponding to the actual image.

- `$BASE64_IMAGE`: A pure Base64-encoded string of the image file (without any prefix).

</div>

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "data:{MIME_TYPE};base64,$BASE64_IMAGE"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "data:{MIME_TYPE};base64,$BASE64_IMAGE"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

#### Anthropic Messages API

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from anthropic import Anthropic

client = Anthropic(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/anthropic"
)

message = client.messages.create(
    model="mimo-v2.6-pro",
    max_tokens=1024,
    system="You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "source": {
                        "type": "base64",
                        "media_type": "{MIME_TYPE}",
                        "data": "$BASE64_IMAGE"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ]
)

print(message.content)
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/anthropic/v1/messages' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "max_tokens": 1024,
    "system": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.",
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    "type": "image",
                    "source": {
                        "type": "base64",
                        "media_type": "{MIME_TYPE}",
                        "data": "$BASE64_IMAGE"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the image"
                }
            ]
        }
    ]
}'
```

  </TabItem>
</Tab>

### Multi-image Input

{/* feishu-style:text-align:left */}
Supports simultaneously passing in public network URLs or Base64-encoded strings of multiple images, and the model can parse the image content and return responses that match the image semantics.

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/image/image_example.png"
                    }
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "data:{MIME_TYPE};base64,$BASE64_IMAGE"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the connections and differences between these two pictures"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/image/image_example.png"
                    }
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "data:{MIME_TYPE};base64,$BASE64_IMAGE"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the connections and differences between these two pictures"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

## Image Restrictions

- Image Formats: JPEG, PNG, GIF, WebP, BMP. 

- Image Size: 

   - When passed in as a URL: single image file size does not exceed 50 MB.

   - When passed in as Base64 encoding: The size of the Base64 encoded string of a single image does not exceed 50 MB. 

- Number of images: When multiple images are passed in, the number of images is limited by the model's context length, and the total number of Tokens for all images and text must be less than the model's context length.

> Note: For calculating image tokens, please refer to [Explanation of Image Token Usage](https://platform.xiaomimimo.com/#/docs/usage-guide/multimodal-understanding/image-understanding?target=explanation-of-image-token-usage-and-scaling-rules). For the model context length, please refer to [Pricing and Rate Limits](https://platform.xiaomimimo.com/#/docs/pricing). 

## Explanation of Image Token Usage and Scaling Rules

{/* feishu-style:text-align:left */}
The calculation rules for images are relatively complex. For Token conversion and scaling rules, please refer to the following code. The estimated results are for reference only, and the actual usage shall be subject to the API response. 

```python
import math
from PIL import Image

PATCH_SIZE = 16
SPATIAL_MERGE_SIZE = 2
TEMPORAL_PATCH_SIZE = 2
IMAGE_MIN_PIXELS = 8192
IMAGE_MAX_PIXELS = 8388608

def calc_image_tokens(image_path: str) -> dict:
    image = Image.open(image_path)
    height = image.height
    width = image.width

    factor = PATCH_SIZE * SPATIAL_MERGE_SIZE  # 32

    h_bar = round(height / factor) * factor
    w_bar = round(width / factor) * factor

    if h_bar * w_bar > IMAGE_MAX_PIXELS:
        beta = math.sqrt((height * width) / IMAGE_MAX_PIXELS)
        h_bar = math.floor(height / beta / factor) * factor
        w_bar = math.floor(width / beta / factor) * factor
    elif h_bar * w_bar < IMAGE_MIN_PIXELS:
        beta = math.sqrt(IMAGE_MIN_PIXELS / (height * width))
        h_bar = math.ceil(height / beta / factor) * factor
        w_bar = math.ceil(width / beta / factor) * factor

    grid_t = 1
    grid_h = h_bar // PATCH_SIZE
    grid_w = w_bar // PATCH_SIZE
    num_tokens = (grid_t * grid_h * grid_w) // (SPATIAL_MERGE_SIZE ** 2)
    return num_tokens

if __name__ == "__main__":
   token = calc_image_tokens(image_path="xxx/test.jpg")
   print(token)
```

## Price

- Billing: Total cost is calculated based on the number of input, input (cache hits), and output tokens; for pricing, please refer to [Pricing and Rate Limits](https://platform.xiaomimimo.com/#/docs/pricing). 

   - The Token consumption of images can be calculated through [Explanation of Image Token Usage](https://platform.xiaomimimo.com/#/docs/usage-guide/multimodal-understanding/image-understanding?target=explanation-of-image-token-usage-and-scaling-rules). The estimated results are for reference only, and the actual usage shall be subject to the API response. 

- View Bill: You can view your bill and usage on the [Billing](https://platform.xiaomimimo.com/#/console/usage) page in the Console. 

## FAQ

### Does it support local file upload?

{/* feishu-style:text-align:left */}
`mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed` and `mimo-v2.5` models don't currently support uploading local image files. For supported upload methods, please refer to [Image Input Method](https://platform.xiaomimimo.com/#/docs/usage-guide/multimodal-understanding/image-understanding?target=image-input-method).


--- DOCUMENT: Audio Understanding ---
URL: https://mimo.mi.com/static/docs/quick-start/usage-guide/multimodal-understanding/audio-understanding.md

# Audio Understanding

{/* feishu-style:text-align:left */}
The audio understanding model can answer based on the audio you provide, supporting both audio URL and Base64 encoding as input methods, and is suitable for scenarios such as audio analysis. 

## Quick Start

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

For preparations such as obtaining an API Key, please refer to [First API Call](https://platform.xiaomimimo.com/#/docs/quick-start/first-api-call).

</div>

{/* feishu-style:text-align:left */}
Quickly experience the audio understanding effect by passing the audio URL into the model. The sample code is as follows.

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": "https://example-files.cnbj1.mi-fds.com/example-files/audio/audio_example.wav"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the audio"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": "https://example-files.cnbj1.mi-fds.com/example-files/audio/audio_example.wav"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the audio"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Response**

```json
{
    "id": "020d7a2c-f8e7-4e5a-aaea-790ba908b51a_4e073cbfa8994cc1832055183ac841e5",
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "message": {
                "content": "The audio features a female voice saying, \"Good morning,\" followed by the question, \"Could you tell me what the weather will be like today?\"",
                "role": "assistant",
                "tool_calls": null,
                "reasoning_content": "- The audio starts with a female voice saying \"Good morning.\"\n- Then she asks, \"Could you tell me what the weather will be like today?\"\n- The audio is very short, just a few seconds.\n- The content is a simple greeting followed by a question about the weather.\n- I'll summarize this clearly."
            }
        }
    ],
    "created": 1790007961,
    "model": "mimo-v2.6-pro",
    "object": "chat.completion",
    "usage": {
        "completion_tokens": 98,
        "prompt_tokens": 86,
        "total_tokens": 184,
        "completion_tokens_details": {
            "reasoning_tokens": 66
        },
        "prompt_tokens_details": {
            "audio_tokens": 25,
            "cached_tokens": 0
        }
    }
}
```

## Supported models

{/* feishu-style:text-align:left */}
Currently, the `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed` and `mimo-v2.5` models are supported.

## Audio Input method

{/* feishu-style:text-align:left */}
Supported audio input methods are as follows:

- Audio URL Input: A publicly accessible audio URL address must be provided.

- Base64 Encoding Input: Convert the audio to a Base64-encoded string before passing it in.

### Audio URL Input

{/* feishu-style:text-align:left */}
Audio files can be directly passed in via a publicly accessible audio URL address, which is suitable for scenarios where the audio files are already stored in a publicly accessible environment. The size of a single audio file cannot exceed 100 MB.

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": "https://example-files.cnbj1.mi-fds.com/example-files/audio/audio_example.wav"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the audio"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": "https://example-files.cnbj1.mi-fds.com/example-files/audio/audio_example.wav"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the audio"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

### Base64 Encoding Input

{/* feishu-style:text-align:left */}
Convert the audio file to a Base64-encoded string and then pass it in, which is suitable for scenarios where the audio file cannot be accessed via a public network URL. The size of the converted Base64-encoded string cannot exceed 50 MB.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Please include the prefix before Base64 encoding: `data:{MIME_TYPE};base64,$BASE64_AUDIO`

- `{MIME_TYPE}`: The MIME type (media type) of the audio, used to identify the audio format, which needs to be replaced with the MIME value corresponding to the actual audio.

- `$BASE64_AUDIO`: A pure Base64-encoded string of the audio file (without any prefix).

</div>

> In the following example, both `{MIME_TYPE}` and `$BASE64_AUDIO` are placeholders, please replace them with actual valid data before running.

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": "data:{MIME_TYPE};base64,$BASE64_AUDIO"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the audio"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": "data:{MIME_TYPE};base64,$BASE64_AUDIO"
                    }
                },
                {
                    "type": "text",
                    "text": "please describe the content of the audio"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

## Audio Restrictions

- Audio Formats: MP3, WAV, FLAC, M4A, OGG.

> Audio Formats variants are numerous, and it cannot be guaranteed that all files can be recognized. Please verify through testing that the files can be recognized normally.

- Audio Size:

- When passed in as a URL: File size does not exceed 100 MB. 

- When passed in as Base64 encoding: The size of the Base64 encoded string of a single audio file does not exceed 50 MB. 

- Number of audios: When multiple audio files are input, the number of audio files is limited by the model's context length, and the total number of tokens for all audio and text must be less than the model's context length.

> Note: For calculating audio tokens, please refer to [Explanation of Audio Token Usage](https://platform.xiaomimimo.com/#/docs/usage-guide/multimodal-understanding/audio-understanding?target=explanation-of-audio-token-usage). For the model context length, please refer to [Pricing and Rate Limits](https://platform.xiaomimimo.com/#/docs/pricing). 

## Explanation of Audio Token Usage

{/* feishu-style:text-align:left */}
For the Token conversion of audio, please refer to the following code. The estimated results are for reference only, and the actual usage is subject to the API response. 

```bash
Total tokens ≈ Audio duration (in seconds, e.g., 10.6 seconds) * 6.25
```

## Price

- Billing: The total cost is calculated based on the number of input, input (cache hits), and output tokens; for pricing, please refer to [Pricing and Rate Limits](https://platform.xiaomimimo.com/#/docs/pricing). 

- Audio Token consumption can be calculated through [Explanation of Audio Token Usage](https://platform.xiaomimimo.com/#/docs/usage-guide/multimodal-understanding/audio-understanding?target=explanation-of-audio-token-usage). The estimated results are for reference only, and the actual usage is subject to the API response. 

- View Bill: You can view your bill and usage on the [Billing](https://platform.xiaomimimo.com/#/console/usage) page in the Console. 

## FAQ

### Does it support local file upload?

{/* feishu-style:text-align:left */}
`mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed` and `mimo-v2.5` models don't currently support uploading local audio files. For supported upload methods, please refer to [Audio Input Method](https://platform.xiaomimimo.com/#/docs/usage-guide/multimodal-understanding/audio-understanding?target=audio-input-method).


--- DOCUMENT: Video Understanding ---
URL: https://mimo.mi.com/static/docs/quick-start/usage-guide/multimodal-understanding/video-understanding.md

# Video Understanding

{/* feishu-style:text-align:left */}
The video understanding model can answer based on the video you provide, supporting both video URL and Base64 encoding as input methods, and is suitable for scenarios such as video analysis. 

## Quick Start

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

For preparations such as obtaining an API Key, please refer to [First API Call](https://platform.xiaomimimo.com/#/docs/quick-start/first-api-call).

</div>

{/* feishu-style:text-align:left */}
Quickly experience the video understanding effect by passing the model through the video URL method. The sample code is as follows.

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "video_url",
                    "video_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/video/video_example.mp4"
                    },
                    "fps": 2,
                    "media_resolution": "default"
                },
                {
                    "type": "text",
                    "text": "please describe the content of the video"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "video_url",
                    "video_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/video/video_example.mp4"
                    },
                    "fps": 2,
                    "media_resolution": "default"
                },
                {
                    "type": "text",
                    "text": "please describe the content of the video"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Response**

```json
{
    "id": "fbfdb524-f4c8-41e6-b0a3-31520f6198c7_0b8b3ccec31e40868b76109308ed9329",
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "message": {
                "content": "The video features a small, white kitten sitting on a polished wooden floor indoors. The scene is brightly lit by natural sunlight streaming in from a window on the left, which is partially covered by a beige curtain. The light casts distinct shadows on the floorboards. To the right, a large green potted plant (resembling a Monstera) is visible in the background. The kitten, which has striking blue eyes, sits facing forward but glances slightly to the side. It briefly opens its mouth, as if meowing or yawning. The overall atmosphere is warm, calm, and sunny.",
                "role": "assistant",
                "tool_calls": null,
                "reasoning_content": "The user wants a description of the video content.\n\n1.  **Analyze the video:**\n    *   **Subject:** A small, white kitten.\n    *   **Setting:** Indoors, near a window. There's a wooden floor. A curtain hangs to the left. A potted plant (Monstera) is visible in the background to the right.\n    *   **Lighting:** Bright natural sunlight is streaming in from the left (likely a window), casting strong shadows and highlights on the floor and the kitten.\n    *   **Action:** The kitten is sitting on the floor. It's looking slightly to its left (viewer's right). It opens its mouth briefly, possibly meowing or yawning, but there's no audio. Then it looks forward again. It's a very short clip, essentially a still shot with minor movement.\n    *   **Details:** Blue eyes, white fur, pink nose. The curtain is beige/white striped. The floor is polished wood.\n\n2.  **Draft the description:**\n    *   Start with the main subject: A small white kitten.\n    *   Describe the location: Sitting on a wooden floor near a window.\n    *   Describe the lighting: Sunlight is streaming in, creating a warm atmosphere.\n    *   Describe the background elements: A curtain on the left, a green plant on the right.\n    *   Describe the kitten's features: White fur, blue eyes.\n    *   Describe the action: The kitten looks around and opens its mouth slightly.\n\n3.  **Refine the description:**\n    *   \"The video features a small white kitten with blue eyes sitting on a polished wooden floor.\"\n    *   \"The setting appears to be a cozy room with bright sunlight streaming in through a window on the left.\"\n    *   \"A beige curtain hangs beside the window.\"\n    *   \"To the right, a green potted plant is visible in the background.\"\n    *   \"The kitten is facing forward but looks slightly to the side, opening its mouth briefly as if meowing or yawning.\"\n    *   \"The lighting creates distinct shadows on the floor.\"\n\n4.  **Final Polish:** Combine into a coherent paragraph. Ensure it covers the visual aspects clearly.\n\n    *   *Draft:* The video shows a white kitten with bright blue eyes sitting on a shiny wooden floor. Sunlight is pouring in from a window on the left, illuminating the kitten and the floor. There is a curtain to the left and a large leafy plant in the background on the right. The kitten opens its mouth slightly and looks around. The scene is calm and well-lit.\n\n    *   *More detailed:* The video captures a serene moment featuring a small, white kitten sitting on a polished wooden floor bathed in natural sunlight. The light streams in from a window on the left, partially covered by a beige curtain, casting long shadows across the floor. To the right, the large green leaves of a monstera plant add a touch of nature to the indoor setting. The kitten, with its striking blue eyes, faces the camera but glances to the side, briefly opening its mouth as if letting out a soft meow or a yawn. The overall atmosphere is warm and peaceful.\n\n5.  **Output generation:** Present the description clearly to the user."
            }
        }
    ],
    "created": 1790008279,
    "model": "mimo-v2.6-pro",
    "object": "chat.completion",
    "usage": {
        "completion_tokens": 807,
        "prompt_tokens": 1260,
        "total_tokens": 2067,
        "completion_tokens_details": {
            "reasoning_tokens": 684
        },
        "prompt_tokens_details": {
            "audio_tokens": 19,
            "cached_tokens": 1216,
            "video_tokens": 1144
        }
    }
}
```

## Supported models

{/* feishu-style:text-align:left */}
Currently, the `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed` and `mimo-v2.5` models are supported.

## Video Input Method

{/* feishu-style:text-align:left */}
Supported video input methods are as follows:

- Video URL Input: A publicly accessible video URL address must be provided. 

- Base64 Encoding Input: Convert the video to a Base64-encoded string before inputting it.

### Video URL Input

{/* feishu-style:text-align:left */}
Videos can be directly passed in via a publicly accessible video URL address, which is suitable for scenarios where the video is already stored in a publicly accessible environment. The size of a single video file cannot exceed 300 MB. 

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "video_url",
                    "video_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/video/video_example.mp4"
                    },
                    "fps": 2,
                    "media_resolution": "default"
                },
                {
                    "type": "text",
                    "text": "please describe the content of the video"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "video_url",
                    "video_url": {
                        "url": "https://example-files.cnbj1.mi-fds.com/example-files/video/video_example.mp4"
                    },
                    "fps": 2,
                    "media_resolution": "default"
                },
                {
                    "type": "text",
                    "text": "please describe the content of the video"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

### Base64 Encoding Input

{/* feishu-style:text-align:left */}
Convert the video file to a Base64-encoded string and then pass it in, which is suitable for scenarios where the video cannot be accessed via a public network URL. The size of the converted Base64-encoded string cannot exceed 50 MB.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Please include the prefix before Base64 encoding: `data:{MIME_TYPE};base64,$BASE64_VIDEO`

- `{MIME_TYPE}`: The MIME type (media type) of the video, used to identify the video format, and needs to be replaced with the MIME value corresponding to the actual video.

- `$BASE64_VIDEO`: Pure Base64-encoded string of the video file (without any prefix).

</div>

> In the following example, both `{MIME_TYPE}` and `$BASE64_VIDEO` are placeholders, please replace them with actual valid data before running.

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.6-pro",
    messages=[
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "video_url",
                    "video_url": {
                        "url": "data:{MIME_TYPE};base64,$BASE64_VIDEO"
                    },
                    "fps": 2,
                    "media_resolution": "default"
                },
                {
                    "type": "text",
                    "text": "please describe the content of the video"
                }
            ]
        }
    ],
    max_completion_tokens=1024
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "video_url",
                    "video_url": {
                        "url": "data:{MIME_TYPE};base64,$BASE64_VIDEO"
                    },
                    "fps": 2,
                    "media_resolution": "default"
                },
                {
                    "type": "text",
                    "text": "please describe the content of the video"
                }
            ]
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
</Tab>

## Instructions for Use

### Video Restrictions

- Video Formats: MP4, MOV, AVI, WMV.

> Video Formats variants are numerous, and it cannot be guaranteed that all files can be recognized. Please verify through testing that the files can be recognized normally.

- Video Size:

   - When passed in as a URL: single video file size does not exceed 300 MB. 

   - When passed in as Base64 encoding: The size of the Base64 encoded string of a single video does not exceed 50 MB. 

- Number of videos: When multiple videos are input, the number of videos is limited by the model's context length, and the total number of tokens for all video and text must be less than the model's context length.

> Note: For calculating video tokens, please refer to [Explanation of Video Token Usage](https://platform.xiaomimimo.com/#/docs/usage-guide/multimodal-understanding/video-understanding?target=explanation-of-video-token-usage). For the model context length, please refer to [Pricing and Rate Limits](https://platform.xiaomimimo.com/#/docs/pricing). 

### Control the fineness of video understanding

{/* feishu-style:text-align:left */}
You can control the granularity of video understanding through the two fields `fps` and `media_resolution` respectively.

1. `fps` is the frame number of images extracted from the video per second, used to control the fineness of understanding the time dimension of the video. The default value is 2, with a range of `[0.1, 10]`.

   - The higher the value, the denser the frame extraction, and the more refined the model's perception of frame changes, movements, and temporal details; 

   - The lower the value, the sparser the frame extraction, the faster the processing speed, and the less Token consumption.

1. `media_resolution` refers to the resolution level of video frames, used to control the visual understanding fineness of a single frame. The default value is `default`. 

   - `default`: The default level, balancing recognition effectiveness and processing efficiency;

   - `max`: The highest resolution level, which enhances the recognition ability for small objects and detailed textures.

## Explanation of Video Token Usage

{/* feishu-style:text-align:left */}
Video tokens are divided into `video_tokens` (visual) and `audio_tokens` (audio). 

- `video_tokens` Calculation please refer to the following code. The estimated results are for reference only, and the actual usage is subject to the API response. 

   ```python
   """
   Estimate the number of tokens consumed by an API call based on video duration and resolution.
   Two parameters control the level of detail:
     - fps: Frames extracted per second. Default 2, range [0.1, 10]. Higher values yield
       finer temporal granularity at the cost of more tokens.
     - media_resolution: Per-frame resolution tier. "default" balances quality and efficiency;
       "max" improves fine-grained detail recognition.
   """
   
   import math
   
   def estimate_video_tokens(
       duration: float,
       width: int,
       height: int,
       fps: float = 2.0,
       media_resolution: str = "default",
       mute: bool = False,
   ) -> int:
       """
       Estimate the token count for a video input.
   
       Args:
           duration:         Video duration in seconds.
           width:            Video width in pixels.
           height:           Video height in pixels.
           fps:              Frame extraction rate. Default 2, range [0.1, 10].
           media_resolution: "default" or "max".
           mute:             If True, audio tokens are excluded.
   
       Returns:
           Estimated total token count.
       """
       # ---- Constants ----
       PATCH, MERGE, T_PATCH = 16, 2, 2
       SPATIAL = PATCH * MERGE                         # 32
       PIX_PER_TOKEN = SPATIAL ** 2                    # 1024
       MAX_TOTAL_TOKENS = 131072
       TOTAL_MAX_PIX = MAX_TOTAL_TOKENS * PIX_PER_TOKEN
       MIN_PIX, MAX_PIX = 8192, 8388608
       MAX_FRAMES = 2048
       DEFAULT_MAX_FRAME_TOKEN = 300
   
       # ---- 1. Number of extracted frames ----
       nframes = math.ceil(duration * fps)
       nframes = min(nframes, MAX_FRAMES)
       nframes = max(math.ceil(nframes / T_PATCH) * T_PATCH, T_PATCH)
   
       # ---- 2. Per-frame pixel budget ----
       max_pix = TOTAL_MAX_PIX * T_PATCH // nframes
       if media_resolution != "max":
           max_pix = min(max_pix, DEFAULT_MAX_FRAME_TOKEN * PIX_PER_TOKEN)
       max_pix = max(MIN_PIX, min(max_pix, MAX_PIX))
   
       # ---- 3. Resolution scaling ----
       h, w = height, width
       if min(h, w) < SPATIAL:
           if h < w:
               w = int(w * SPATIAL / h); h = SPATIAL
           else:
               h = int(h * SPATIAL / w); w = SPATIAL
       h_bar = round(h / SPATIAL) * SPATIAL
       w_bar = round(w / SPATIAL) * SPATIAL
       if h_bar * w_bar > max_pix:
           beta = math.sqrt(h * w / max_pix)
           h_bar = math.floor(h / beta / SPATIAL) * SPATIAL
           w_bar = math.floor(w / beta / SPATIAL) * SPATIAL
       elif h_bar * w_bar < MIN_PIX:
           beta = math.sqrt(MIN_PIX / (h * w))
           h_bar = math.ceil(h * beta / SPATIAL) * SPATIAL
           w_bar = math.ceil(w * beta / SPATIAL) * SPATIAL
   
       # ---- 4. Token calculation ----
       grids = nframes // T_PATCH                       # temporal grid count
       tokens_per_grid = (h_bar // PATCH) * (w_bar // PATCH) // (MERGE ** 2)
       vision = grids * tokens_per_grid
       timestamps = grids * (5 if fps > 2 else 3)       # timestamp text tokens
       special = grids * 2 + 2                           # special markers
   
       # ---- 5. Audio tokens ----
       audio = 0
       if not mute:
           spec_len = int(duration * 24000) // 240 + 1
           t = (spec_len - 1) // 2 + 1
           t = t // 2 + int(t % 2 != 0)
           audio = math.ceil(t / 4) + 2                 # +2 for audio special tokens
   
       return vision + timestamps + special + audio
   
   # ============ Example ============
   if __name__ == "__main__":
       # A 1080p, 60-second video
       tokens = estimate_video_tokens(duration=60, width=1920, height=1080)
       print(f"Default params (fps=2, default): {tokens:,} tokens")
   
       tokens = estimate_video_tokens(duration=60, width=1920, height=1080, fps=5)
       print(f"High frame rate (fps=5, default): {tokens:,} tokens")
   
       tokens = estimate_video_tokens(duration=60, width=1920, height=1080, media_resolution="max")
       print(f"High resolution (fps=2, max):     {tokens:,} tokens")
   
       tokens = estimate_video_tokens(duration=60, width=1920, height=1080, mute=True)
       print(f"Muted           (fps=2, mute):    {tokens:,} tokens")
   ```

- `audio_tokens` Calculation please refer to the following code. The estimated results are for reference only, and the actual usage is subject to the API response. 

   ```bash
   Total tokens ≈ Audio duration (in seconds) * 6.25
   ```

## Price

- Billing: The total cost is calculated based on the number of input, input (cache hits), and output tokens; for pricing, please refer to [Pricing and Rate Limits](https://platform.xiaomimimo.com/#/docs/pricing). 

   - Video Token consumption can be calculated through [Explanation of Video Token Usage](https://platform.xiaomimimo.com/#/docs/usage-guide/multimodal-understanding/video-understanding?target=explanation-of-video-token-usage). The estimated results are for reference only, and the actual usage is subject to the API response. 

- View Bill: You can view your bill and usage on the [Billing](https://platform.xiaomimimo.com/#/console/usage) page in the Console. 

## FAQ

### Does it support local file upload?

{/* feishu-style:text-align:left */}
`mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed` and `mimo-v2.5` models don't currently support uploading local video files. For supported upload methods, please refer to [Video Input Method](https://platform.xiaomimimo.com/#/docs/usage-guide/multimodal-understanding/video-understanding?target=video-input-method).


--- DOCUMENT: Speech Recognition（MiMo-V2.5-ASR） ---
URL: https://mimo.mi.com/static/docs/quick-start/usage-guide/audio/Speech-Recognition.md

# Speech Recognition (MiMo-V2.5-ASR)

{/* feishu-style:text-align:left */}
Speech recognition converts input audio into text output, suitable for meeting transcription, lyrics recognition, dialect transcription, noisy environment recordings, and more. You can improve recognition accuracy by specifying language parameters.

{/* feishu-style:text-align:left */}
**Core Capabilities**

- **Broad Language and Dialect Coverage**: Supports bilingual Chinese-English recognition with automatic language detection. Natively recognizes Cantonese, Wu, Minnan, Sichuan, and other Chinese dialects.

- **Robust in Complex Scenarios**: Maintains stable recognition in noisy environments, far-field pickup, and multi-speaker overlapping conversations. Also supports lyrics transcription with background music.

- **Precise Handling of Specialized Content**: Accurately recognizes knowledge-intensive content such as classical poetry, technical terminology, proper nouns, and place names. Automatically generates punctuation without post-processing.

## Supported Models

{/* feishu-style:text-align:left */}
Currently, only the `mimo-v2.5-asr` model is supported.

## Prerequisites

{/* feishu-style:text-align:left */}
For API Key setup and other prerequisites, please refer to [First API Call](https://platform.xiaomimimo.com/#/docs/quick-start/first-api-call).

## Supported Audio Formats

{/* feishu-style:text-align:left */}
Currently, only `wav` and `mp3` audio sample files are supported. Before passing audio to the API, convert the file to a Base64 encoded string. The encoded string size must not exceed 10 MB. Currently, two audio input methods are supported:

<Tab>
  <TabItem label={`Data URL`}>

```json
"input_audio": {
    "data": "data:{MIME_TYPE};base64,$BASE64_AUDIO"
}
```

  </TabItem>
  <TabItem label={`Pure Base64 Encoding`}>

{/* feishu-style:text-align:left */}
When passing audio in pure Base64 encoding, the `format` field must be passed simultaneously to specify the audio format.

```json
"input_audio": {
    "data": "$BASE64_AUDIO",
    "format": "{format}"
}
```

  </TabItem>
</Tab>

{/* feishu-style:text-align:left */}
**Supported formats and their MIME types:** 

<table>
<colgroup>
<col style="width: 50%" />
<col style="width: 50%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Format</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">MIME Type</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">wav</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`audio/wav`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mp3</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`audio/mpeg` or `audio/mp3`</span></td>
</tr>
</tbody>
</table>

## Code Sample

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Set `asr_options.language` to specify the language. Auto detection will be used if this parameter is not configured. Manual specification is recommended when the language is confirmed to improve recognition performance. Supported values: `auto`, `zh`, `en`.

</div>

### Non-streaming Call

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
import base64
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

# Replace with the actual local file path
with open("audio_file.wav", "rb") as f:
    audio_bytes = f.read()
audio_base64 = base64.b64encode(audio_bytes).decode("utf-8")

completion = client.chat.completions.create(
    model="mimo-v2.5-asr",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": f"data:audio/wav;base64,{audio_base64}"
                    }
                }
            ]
        }
    ],
    extra_body={
        "asr_options": {
            "language": "en"
        }
    }
)

print(completion.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "mimo-v2.5-asr",
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": "data:{MIME_TYPE};base64,$BASE64_AUDIO"
                    }
                }
            ]
        }
    ],
    "asr_options": {
        "language": "en"
    }
}'
```

  </TabItem>
</Tab>

### Streaming Call

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
import base64
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

# Replace with the actual local file path
with open("audio_file.wav", "rb") as f:
    audio_bytes = f.read()
audio_base64 = base64.b64encode(audio_bytes).decode("utf-8")

completion = client.chat.completions.create(
    model="mimo-v2.5-asr",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": f"data:audio/wav;base64,{audio_base64}"
                    }
                }
            ]
        }
    ],
    extra_body={
        "asr_options": {
            "language": "auto"
        }
    },
    stream=True
)

for chunk in completion:
    print(chunk.model_dump_json())
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "mimo-v2.5-asr",
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    "type": "input_audio",
                    "input_audio": {
                        "data": "data:{MIME_TYPE};base64,$BASE64_AUDIO"
                    }
                }
            ]
        }
    ],
    "asr_options": {
        "language": "auto"
    },
    "stream": true
}'
```

  </TabItem>
</Tab>

## Price

- Billing: Please refer to [Pay‑As‑You‑Go API](https://platform.xiaomimimo.com/#/docs/pricing).

- View Bill: You can view your usage on the [Billing](https://platform.xiaomimimo.com/#/console/usage) page in the Console.


--- DOCUMENT: Speech Synthesis (MiMo-V2.5-TTS Series) ---
URL: https://mimo.mi.com/static/docs/quick-start/usage-guide/audio/speech-synthesis-v2.5.md

# Speech Synthesis (MiMo-V2.5-TTS Series)

{/* feishu-style:text-align:left */}
Speech Synthesis (Text-to-Speech) supports automatically converting input text into natural and fluent speech output. You can generate natural and vivid speech content by configuring parameters such as speech style and voice.

{/* feishu-style:text-align:left */}
**Core Capabilities**

- **Out-of-the-box built-in voices:**  A variety of high-quality built-in voices are available for quick use without additional configuration.

- **Voice design and cloning:**  Supports voice design via text description, or replication of arbitrary voices based on audio samples.

- **Diverse speech styles:**  Supports control over speed, emotion, role-play, dialects and other styles, for more vivid and natural speech expression.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Low-latency streaming output for `mimo-v2.5-tts` is now available. The streaming interface has been restored and returns responses in real time.

</div>

## List of Supported Models

{/* feishu-style:text-align:left */}
Currently, three models of the MiMo-V2.5-TTS series are supported, and the model list is as follows:

<table>
<colgroup>
<col style="width: 23.87%" />
<col style="width: 22.4%" />
<col style="width: 23.58%" />
<col style="width: 30.150000000000006%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Model ID</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Function</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Voice</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Precautions </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Use built-in high-quality voices for speech synthesis </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Use the high-quality voices from the built-in voices list </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Supports singing mode, does not support voice design and voice cloning</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts-voicedesign`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Customize voice through text description</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Automatically generate voices from text descriptions, without requiring presets or audio samples</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Does not support singing mode, built-in voices, or voice cloning</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts-voiceclone`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Replicate any voice from audio samples</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Precisely replicate voices from audio samples to enable speech synthesis of any voice</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Does not support singing mode, built-in voices, or voice design</span></td>
</tr>
</tbody>
</table>

## Preparation

{/* feishu-style:text-align:left */}
For preparations such as obtaining API Key, please refer to [ First API Call ](https://platform.xiaomimimo.com/#/docs/quick-start/first-api-call). 

## General Precautions

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

**Call Rules**

- The target text for speech synthesis must be filled in the `role` of `assistant` message and cannot be placed in the `user` role message. 

- `user` role messages are optional parameters, and instructions can be passed in to adjust the tone and style of speech synthesis, or they can be conversation history (message content will not appear in synthesized speech). When using the `mimo-v2.5-tts-voicedesign` model, they are required parameters. 

- When using streaming calls, please specify the format of the output audio as `pcm16`, so that it can be spliced into a complete audio. For splicing examples, please refer to the Python calling methods in each chapter.

</div>

## Style Control 

{/* feishu-style:text-align:left */}
The instruction-following ability of the model is sufficient to cover the following complex controls (a single natural language instruction is sufficient to take effect): 

- **Multi-style Switching**: A single character completes the style transition from *announcement → whisper → roar* within the same voice segment, with a natural and unobtrusive transition.

- **Multi-emotion Mixing**: Supports complex emotions such as "repressed anger", "smile with a sob", "gentle but tired", "gentleness in mania", etc., rather than only allowing the selection of a single emotion.

- **Multi-granularity control**: From *paragraph level* (overall tone) → *sentence level* (rhythm) → *word level* (stress) → *character granularity* (choking, dragging, or breathy sound of a specific character), all can be specified in the instruction.

{/* feishu-style:text-align:left */}
We currently offer two control methods: **natural language control** and **tag control** . The placement of the content for both methods in ` messages ` is different: 

- **Natural Language Control** → Placed in `role: user`'s `content`

- **Audio Tag Control**  → Placed in ` role: assistant ` 's ` content `

### Natural Language Control

{/* feishu-style:text-align:left */}
Through natural language description, enable the model to understand and generate speech in the corresponding style. **The content is placed in the** `messages` **field of** `role: user` **in the** `content` **field.**   You can directly describe the desired speech style in a single sentence. 

{/* feishu-style:text-align:left */}
**Example:** 

> Report good news to the leader in a brisk and upbeat tone, speaking at a slightly faster pace, with the uncontrollable excitement and a touch of pride after learning the results, and a bright and energetic voice. 

> Looking at the results of the just-solved difficult problem, couldn't help exclaiming in a self-satisfied and overjoyed manner, with a high-pitched and bright voice, a relatively fast speaking speed, and a tone full of confidence and disbelief. 

> With a bright and lively teenage voice, carrying the pride and playfulness after a successful prank, speaking at a relatively fast pace with light enunciation, and the tone slightly rising when emphasizing the bet. 

{/* feishu-style:text-align:left */}
On this basis, we also support a more complex and refined **director mode** — just like writing a script for actors, comprehensively depicting characters and voices from the three dimensions of **character, scene, and guidance**, based on which the model can generate more layered and performative voices.

- **[Character]**  Clearly describe the character's identity, personality traits, physical appearance and speaking habits.

- **[Scene]**  Describe what is happening at this moment, who you are talking to, and what emotional state you are in. The more specific the better — time, location, event, and the other person's reaction can all be included.

- **[Guidance]**  Similar to a director giving acting instructions to an actor: speaking speed, breath control, pauses, accents, resonance position, timbre texture, and emotional fluctuations. It can be written in detail, and the model will act according to these "stage directions". 

{/* feishu-style:text-align:left */}
**Example:** 

```python
Role: The current head of the century-old noble Cen family. Since birth, she was adopted and raised by the gatekeeper of the ancestral temple, molded into a flawless, emotionless family totem. She has long lived in seclusion and has a strong sense of class alienation towards others.

Scene: In the shadows of the ancestral hall, she watches the man who has broken through the security cordon at all costs to find her and attempts to elope with her. She will use the coldest and most rigid class barriers to strangle both the other person and the feelings that have just sprouted but are enough to start a prairie fire within herself.

Guidance:
A cold, languid yet extremely imposing deep-voiced mature woman. Her vocal tract is very relaxed, without any sign of tension, yet exuding a bone-chilling sense of oppression.

- Speed and Pauses: Extremely slow, with each word rolling on the tip of her tongue before being uttered, carrying the casual arrogance of a superior. There are extremely long, unsettling pauses between sentences.
- Breathiness and Full Voice: Most of the time, her voice has no obvious pitch fluctuations, with a heavy and hard full voice, like a calm yet cold undercurrent. However, a very slight breathy sound must be added at certain final sounds (such as "sincerity") to reveal a hint of weariness and longing that even she herself is unaware of.
- Articulation Texture: The mixed use of literary and colloquial words bears the traces of the old era, with labiodental sounds pronounced extremely lightly but extremely clearly (such as "collision" and "cheap"), making her speech both elegant and sharp, hitting home with every word.
```

{/* feishu-style:text-align:left */}
Director Mode is suitable for scenarios with high requirements for voice performance, such as character voiceovers, film-level content generation, etc.

### Audio Tag Control

{/* feishu-style:text-align:left */}
By embedding style tags and audio tags in the text, fine-grained control over speech can be directly achieved. The overall style tag comes at the beginning, and fine-grained control tags can be inserted in the middle. **All tag control content is placed in the** ` messages ` **of the** ` role: assistant ` ` content `  **field.**  

{/* feishu-style:text-align:left */}
Add a **start** `(style)` tag to the target text to specify the pronunciation style of the voice. Multiple styles can be set simultaneously by placing multiple style names within the same pair of parentheses, with no restrictions on the delimiter.

{/* feishu-style:text-align:left */}
**Supported bracket formats:**  Half-width `()`, full-width `（）`, or `[]` can be used.

{/* feishu-style:text-align:left */}
**Format Example:**  `(Style 1 Style 2)Content to be Synthesized`

{/* feishu-style:text-align:left */}
The following are some recommended styles, and custom styles not listed are also supported. 

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

To experience a better singing style, you must add the `(唱歌)` tag at the very beginning of the target text, with the format: `(唱歌)lyrics`. `Lyrics` are recommended to be in Chinese for better synthesis results. The tag supports equivalent values: `唱歌`, `sing`, `singing`. Entering any of them will yield the same result.

</div>

<table>
<colgroup>
<col style="width: 18.57%" />
<col style="width: 81.43%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Style Type**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Style Example** </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Basic Emotions</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Happy / Sad / Angry / Fearful / Amazed / Excited / Wronged / Calm / Indifferent*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Complex Emotions</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Melancholy / Relieved / Helpless / Guilty / Relieved / Jealous / Tired / Apprehensive / Emotional*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Overall tone</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Gentle / Cold / Lively / Serious / Lazy / Playful / Deep / Capable / Sharp*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Timbre Positioning</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Magnetic / Mellow / Clear / Ethereal / Innocent / Old / Sweet / Hoarse / Elegant*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Character Tone</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Clamp voice / Big Sister voice / Shota voice / Uncle voice / Taiwanese accent*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Dialect</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Northeast dialect / Sichuan dialect / Henan dialect / Cantonese*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Role-playing</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Sun Wukong / Lin Daiyu*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Singing </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*singing*</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
**Example:** 
- `(Sighing)After all these years, when I walked down that street again, a part of my heart suddenly felt empty.`
- `(Lazy)Let me sleep for five more minutes... just five minutes, really, for the last time.`
- `(Magnetic)The night is already deep, but the city is still breathing. I'm the one accompanying you tonight. Welcome to listen to <Midnight Radio>.`
- `(Northeastern dialect)Oh my goodness, it's so cold today! You know that wind, it's whistling like a knife, cutting into your face!`
- `(Cantonese)This is really amazing! Once you've tasted it, you won't forget!`
- `(singing)Forgive me for my unruly and unrestrained love for freedom throughout my life, and I'm also afraid that one day I'll fall, Oh no. Abandoning ideals, anyone can do it, so how could I be afraid that one day it'll only be you and me.`

{/* feishu-style:text-align:left */}
On this basis, we also support inserting `[audio tag]` at any position in the text. Through the [audio tag], you can perform fine-grained control over the sound, precisely adjusting tone, mood, and expression style—whether it's a whisper, a hearty laugh, or a little complaint with a touch of emotion. You can also flexibly insert breathing sounds, pauses, coughs, etc., all of which can be easily achieved. The speaking speed can also be flexibly adjusted, allowing each sentence to have its proper rhythm.

<table>
<colgroup>
<col style="width: 23.83%" />
<col style="width: 76.17%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Style Type**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Style Example** </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Speech Rate and Rhythm</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Inhale / Take a deep breath / Sigh / Let out a long sigh / Pant / Hold one's breath*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Emotional State</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*nervous / scared / excited / tired / wronged / coquettish / guilty / shocked / impatient*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Speech Features</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Trembling / Voice trembling / Pitch change / Cracked voice / Nasal voice / Breathiness / Hoarseness*</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Laughing and crying tone</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">*Smile / Chuckle / Laugh out loud / Sneer / Sob / Whimper / Choke / Wail*</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
**Example:** 
- (nervously, takes a deep breath) Hoo... Calm down, calm down. It's just an interview... (speaking faster, muttering) I've rehearsed my self-introduction fifty times, it should be okay. Come on, you can do it... (softly) Oh, is my tie crooked?
- (extremely exhausted, listless) Master... wake me up when we get there... (sighs deeply) I'll take a little nap first. This overtime has made me feel like my soul is about to scatter. 
- If I had... (pauses for a moment) even if I had persisted for just one more second, would the outcome have been different? (forced smile) Oh, there are no "what ifs" anymore. 
- (Rapid breathing due to the cold) Hoo—hoo—This, this snow in the Greater Khingan Mountains... (cough) It can literally freeze one's bones... Don't, don't stop, keep moving, move quickly. 
- (raising voice and shouting) Sister! This fish is fresh! Just caught this morning! Hey! You there, stop rummaging around! If you crush it, you'll have to pay for it! 

## Speech Synthesis Using Built-in Voices

- It comes with multiple high-quality voices and can be used directly without additional configuration. Currently, only the `mimo-v2.5-tts` model is supported

- Supports controlling the style of synthetic speech by passing natural language instructions in the user message

- Supports controlling the style of synthesized speech through audio tags 

### Built-in Voice List

{/* feishu-style:text-align:left */}
When in use, you can set the preset timbre in `{"audio": {"voice": "mimo_default"}}`.

<table>
<colgroup>
<col style="width: 11.51%" />
<col style="width: 13.41%" />
<col style="width: 11.93%" />
<col style="width: 63.15%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Voice Name**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Voice ID** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Language</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Gender</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">MiMo-默认</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo_default</span></td>
<td colspan="2"><span style="display: inline-block; width: 100%; text-align: left;">It varies depending on the deployed cluster. The default for the China cluster is `冰糖`, and the default for other clusters is `Mia`</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">冰糖</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">冰糖</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Chinese</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Female </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">茉莉</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">茉莉</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Chinese</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Female </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">苏打</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">苏打</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Chinese</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Male </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">白桦</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">白桦</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Chinese</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Male </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Mia</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Mia</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">English</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Female </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Chloe</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Chloe</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">English</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Female </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Milo</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Milo</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">English</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Male </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Dean</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Dean</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">English</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Male </span></td>
</tr>
</tbody>
</table>

### Code Sample

#### Non-streaming Call

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI
import base64

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.5-tts",
    messages=[
        {
            "role": "user",
            "content": "Bright, bouncy, slightly sing-song tone — like you're bursting with good news you can barely hold in. Fast pace, rising pitch at the end."
        },
        {
            "role": "assistant",
            "content": "Hey boss — guess what, guess what? I just got the results back and I actually passed! Not just passed, I got a distinction! I know, I know — you told me I was cutting it close, but hey, here we are. Drinks are on me tonight, okay?"
        }
    ],
    audio={
        "format": "wav",
        "voice": "Chloe"
    }
)

message = completion.choices[0].message
audio_bytes = base64.b64decode(message.audio.data)
with open("audio_file.wav", "wb") as f:
    f.write(audio_bytes)
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "mimo-v2.5-tts",
    "messages": [
        {
            "role": "user",
            "content": "Bright, bouncy, slightly sing-song tone — like you are bursting with good news you can barely hold in. Fast pace, rising pitch at the end."
        },
        {
            "role": "assistant",
            "content": "Hey boss — guess what, guess what? I just got the results back and I actually passed! Not just passed, I got a distinction! I know, I know — you told me I was cutting it close, but hey, here we are. Drinks are on me tonight, okay?"
        }
    ],
    "audio": {
        "format": "wav",
        "voice": "Chloe"
    }
}'
```

  </TabItem>
</Tab>

#### Streaming Call

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Low-latency streaming output for `mimo-v2.5-tts` is now available. The streaming interface has been restored and returns responses in real time.

</div>

<Tab>
  <TabItem label={`Python SDK`}>

```python
import base64
import os
import numpy as np
import soundfile as sf
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.5-tts",
    messages=[
        {
            "role": "user",
            "content": "Bright, bouncy, slightly sing-song tone — like you're bursting with good news you can barely hold in. Fast pace, rising pitch at the end."
        },
        {
            "role": "assistant",
            "content": "Hey boss — guess what, guess what? I just got the results back and I actually passed! Not just passed, I got a distinction! I know, I know — you told me I was cutting it close, but hey, here we are. Drinks are on me tonight, okay?"
        }
    ],
    audio={
        "format": "pcm16",
        "voice": "Chloe"
    },
    stream=True
)

# 24kHz PCM16LE mono audio
collected_chunks: np.ndarray = np.array([], dtype=np.float32)

for chunk in completion:
    if not chunk.choices:
        continue
    delta = chunk.choices[0].delta
    audio = getattr(delta, "audio", None)

    if audio is not None:
        assert isinstance(audio, dict), f"Expected audio to be a dict, got {type(audio)}"
        pcm_bytes = base64.b64decode(audio["data"])
        np_pcm = np.frombuffer(pcm_bytes, dtype=np.int16).astype(np.float32) / 32768.0
        collected_chunks = np.concatenate((collected_chunks, np_pcm))
        print(f"Received audio chunk of size {len(pcm_bytes)} bytes")

# Save the collected audio to a file
os.makedirs("tmp", exist_ok=True)
sf.write("tmp/output.wav", collected_chunks, samplerate=24000)
print("Audio saved to tmp/output.wav")
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "mimo-v2.5-tts",
    "messages": [
        {
            "role": "user",
            "content": "Bright, bouncy, slightly sing-song tone — like you are bursting with good news you can barely hold in. Fast pace, rising pitch at the end."
        },
        {
            "role": "assistant",
            "content": "Hey boss — guess what, guess what? I just got the results back and I actually passed! Not just passed, I got a distinction! I know, I know — you told me I was cutting it close, but hey, here we are. Drinks are on me tonight, okay?"
        }
    ],
    "audio": {
        "format": "pcm16",
        "voice": "Chloe"
    },
    "stream": true
}'
```

  </TabItem>
</Tab>

## Speech Synthesis Using Voice Design

{/* feishu-style:text-align:left */}
There is no need to provide an audio file. Simply add voice description text to the message with the role of `user`, and a customized voice can be generated. Currently, only the `mimo-v2.5-tts-voicedesign` model is supported.

### How to Write a Good Voice Design Prompt

{/* feishu-style:text-align:left */}
When using the `mimo-v2.5-tts-voicedesign` model, the text in the `user` message is the voice design description. The more specific and vivid the description, the closer the generated voice will be to the expected one.

#### Key Dimension

{/* feishu-style:text-align:left */}
A good voice description usually covers the following multiple dimensions (not necessarily comprehensive):

<table>
<colgroup>
<col style="width: 25.19%" />
<col style="width: 74.81%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Dimension</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Example</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Gender and Age</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">"young woman in her mid-20s", "middle-aged man in his 50s"</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Voice / Texture</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">"deep and gravelly", "silky, mellow, and magnetic"</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Mood / Tone</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">"warm and confident", "gentle but with a hint of weariness"</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Speech speed / Rhythm</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">"slow and deliberate", "speaking at an extremely fast pace, like a machine gun."</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
The following dimensions can be optionally added to increase richness:

- **Role / Character**: narrator, podcast host, storyteller, late-night radio DJ

- **Speaking style**: casual and colloquial, seriously, lowering one's voice as if plotting

- **Scene description**: narrating a nature documentary, during a roadshow for investors

- **Era reference**: 1940s film noir, dubbed voices of translated films from the 1980s

#### Writing Suggestions

{/* feishu-style:text-align:left */}
**Concise descriptive** -- quickly outline the sound profile using keywords or a single sentence

```bash
Heavy Russian accent, gruff middle-aged male, blunt and matter-of-fact.
```

{/* feishu-style:text-align:left */}
**Professional Descriptive** -- Three-dimensional portrayal of sound through scenarios, character design, or multi-dimensional details

```bash
Young female, extreme close-up with a binaural, ear-to-ear ASMR feel. Audible breathing, subtle swallowing, and soft natural lip sounds. She speaks very slowly, creating a deeply relaxing and immersive experience.
```

```json
An elderly gentleman, speaking Mandarin with a northern accent, his speech slow and steady, his voice slightly hoarse and weathered, as if an old and seasoned grandfather were telling a story, full of the wisdom of years.
```

#### Precautions

- **Length**: 1-4 sentences are sufficient; there's no need to write a long text. Clearly describing the core features is more important than piling up dimensions

- **Avoid conflicts**: Do not simultaneously request contradictory characteristics (e.g., "innocent childish voice + CEO aura")

- **Avoid using audio quality effect terms**: Do not write descriptions related to post-processing such as reverb, echo, EQ, compression, etc

- **Avoid vague words**: Do not use descriptions lacking specific references such as "ordinary," "normal," or "foreign"

- **Both Chinese and English are supported**: the model supports both Chinese and English voice timbre descriptions, so choose the language in which you can express most precisely

- **Synthetic text should match the voice tone**: The synthetic text in the `assistant` message should match the voice tone description to achieve the best results. For example, pair a goodnight monologue with a "gentle and soothing female voice" instead of a passionate sports commentary. It is recommended to use LLM to automatically generate matching synthetic text based on your voice tone description; on the Studio page, you can directly click the "Generate Text" button after entering the voice tone description.

### Code Sample

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

`mimo-v2.5-tts-voicedesign` supports the optional parameter `optimize_text_preview` to control whether the target broadcast text is intelligently polished. When set to `true`, the `assistant` role message can be omitted.

</div>

#### Non-streaming Call

<Tab>
  <TabItem label={`Python SDK`}>

```python
import os
from openai import OpenAI
import base64

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.5-tts-voicedesign",
    messages=[
        {
            "role": "user",
            "content": "Give me a young male tone."
        },
        {
            "role": "assistant",
            "content": "Yes, I had a sandwich."
        }
    ],
    audio={
        "format": "wav",
        "optimize_text_preview": True
    }
)

message = completion.choices[0].message
audio_bytes = base64.b64decode(message.audio.data)
with open("audio_file.wav", "wb") as f:
    f.write(audio_bytes)
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "mimo-v2.5-tts-voicedesign",
    "messages": [
        {
            "role": "user",
            "content": "Give me a young male tone."
        },
        {
            "role": "assistant",
            "content": "Yes, I had a sandwich."
        }
    ],
    "audio": {
        "format": "wav",
        "optimize_text_preview": true
    }
}'
```

  </TabItem>
</Tab>

#### Streaming Call

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

**Note**

- The low-latency streaming output feature of `mimo-v2.5-tts-voicedesign` is not yet available. If you have relevant requirements, please follow the upcoming feature updates.

- The streaming call interface is currently downgraded to compatibility mode, and only **returns the results once in streaming format after all inferences are completed.**

</div>

<Tab>
  <TabItem label={`Python SDK`}>

```python
import base64
import os
import numpy as np
import soundfile as sf
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1"
)

completion = client.chat.completions.create(
    model="mimo-v2.5-tts-voicedesign",
    messages=[
        {
            "role": "user",
            "content": "Give me a young male tone."
        },
        {
            "role": "assistant",
            "content": "You are UN-BE-LIEVABLE! I am sooooo done with your constant lies. GET. OUT!"
        }
    ],
    audio={
        "format": "pcm16",
        "optimize_text_preview": True
    },
    stream=True
)

# 24kHz PCM16LE mono audio
collected_chunks: np.ndarray = np.array([], dtype=np.float32)

for chunk in completion:
    if not chunk.choices:
        continue
    delta = chunk.choices[0].delta
    audio = getattr(delta, "audio", None)

    if audio is not None:
        assert isinstance(audio, dict), f"Expected audio to be a dict, got {type(audio)}"
        pcm_bytes = base64.b64decode(audio["data"])
        np_pcm = np.frombuffer(pcm_bytes, dtype=np.int16).astype(np.float32) / 32768.0
        collected_chunks = np.concatenate((collected_chunks, np_pcm))
        print(f"Received audio chunk of size {len(pcm_bytes)} bytes")

# Save the collected audio to a file
os.makedirs("tmp", exist_ok=True)
sf.write("tmp/output.wav", collected_chunks, samplerate=24000)
print("Audio saved to tmp/output.wav")
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "mimo-v2.5-tts-voicedesign",
    "messages": [
        {
            "role": "user",
            "content": "Give me a young male tone."
        },
        {
            "role": "assistant",
            "content": "You are UN-BE-LIEVABLE! I am sooooo done with your constant lies. GET. OUT!"
        }
    ],
    "audio": {
        "format": "pcm16",
        "optimize_text_preview": true
    },
    "stream": true
}'
```

  </TabItem>
</Tab>

## Speech Synthesis Using Voice Cloning

- By passing in audio samples, you can accurately replicate the target timbre and generate speech. Currently, only the `mimo-v2.5-tts-voiceclone` model is supported 

- Supports controlling the style of synthetic speech by passing natural language instructions in the user message

- Supports controlling the style of synthesized speech through audio tags 

### Code Sample

{/* feishu-style:text-align:left */}
Convert the audio file sample to a Base64-encoded string and then pass it in. The size of the converted Base64-encoded string cannot exceed 10 MB, and currently only `mp3` and `wav` format audio sample files are supported.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Please include the prefix before Base64 encoding: `data:{MIME_TYPE};base64,$BASE64_AUDIO`

- `{MIME_TYPE}`: The MIME type (media type) of the audio, used to identify the audio format, needs to be replaced with the MIME value corresponding to the actual audio. The values here can be: ` audio/mpeg ` (or ` audio/mp3 `), ` audio/wav `. 

- `$BASE64_AUDIO`: A pure Base64-encoded string of the audio file (without any prefix).

</div>

> In the following example, both `{MIME_TYPE}` and `$BASE64_AUDIO` are placeholders, please replace them with actual valid data before running.

#### Non-streaming Call

<Tab>
  <TabItem label={`Python SDK`}>

```python
import base64
import os

from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1",
)

with open("voice.mp3", "rb") as f:
    voice_bytes = f.read()
voice_base64 = base64.b64encode(voice_bytes).decode("utf-8")

completion = client.chat.completions.create(
    model="mimo-v2.5-tts-voiceclone",
    messages=[
        {
            "role": "user",
            "content": ""
        },
        {
            "role": "assistant", 
            "content": "Yes, I had a sandwich."
        }
    ],
    audio={
        "format": "wav",
        "voice": f"data:audio/mpeg;base64,{voice_base64}"
    }
)

message = completion.choices[0].message
audio_bytes = base64.b64decode(message.audio.data)
with open("audio_file.wav", "wb") as f:
    f.write(audio_bytes)
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "mimo-v2.5-tts-voiceclone",
    "messages": [
        {
            "role": "user",
            "content": ""
        },
        {
            "role": "assistant",
            "content": "Yes, I had a sandwich."
        }
    ],
    "audio": {
        "format": "wav",
        "voice": "data:{MIME_TYPE};base64,$BASE64_AUDIO"
    }
}'
```

  </TabItem>
</Tab>

#### Streaming Call

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

**Note**

- The low-latency streaming output feature of `mimo-v2.5-tts-voiceclone` is not yet available. If you have relevant requirements, please follow the upcoming feature updates.

- The streaming call interface is currently downgraded to compatibility mode, and only **returns the results oncein streaming formatafter all inferences are completed.**

</div>

<Tab>
  <TabItem label={`Python SDK`}>

```python
import base64
import os

import numpy as np
import soundfile as sf
from openai import OpenAI

client = OpenAI(
    api_key=os.environ.get("MIMO_API_KEY"),
    base_url="https://api.xiaomimimo.com/v1",
)

with open("voice.mp3", "rb") as f:
    voice_bytes = f.read()
voice_base64 = base64.b64encode(voice_bytes).decode("utf-8")

completion = client.chat.completions.create(
    model="mimo-v2.5-tts-voiceclone",
    messages=[
        {
            "role": "user",
            "content": ""
        },
        {
            "role": "assistant", 
            "content": "Yes, I had a sandwich."
        }
    ],
    audio={
        "format": "wav",
        "voice": f"data:audio/mpeg;base64,{voice_base64}",
    },
    stream=True
)

# 24kHz PCM16LE mono audio
collected_chunks: np.ndarray = np.array([], dtype=np.float32)

for chunk in completion:
    if not chunk.choices:
        continue
    delta = chunk.choices[0].delta
    audio = getattr(delta, "audio", None)

    if audio is not None:
        assert isinstance(audio, dict), (
            f"Expected audio to be a dict, got {type(audio)}"
        )
        pcm_bytes = base64.b64decode(audio["data"])
        np_pcm = np.frombuffer(pcm_bytes, dtype=np.int16).astype(np.float32) / 32768.0
        collected_chunks = np.concatenate((collected_chunks, np_pcm))
        print(f"Received audio chunk of size {len(pcm_bytes)} bytes")

# Save the collected audio to a file
os.makedirs("tmp", exist_ok=True)
sf.write("tmp/output.wav", collected_chunks, samplerate=24000)
print("Audio saved to tmp/output.wav")
```

  </TabItem>
  <TabItem label={`Curl`}>

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header 'Content-Type: application/json' \
--data-raw '{
    "model": "mimo-v2.5-tts-voiceclone",
    "messages": [
        {
            "role": "user",
            "content": ""
        },
        {
            "role": "assistant",
            "content": "You are UN-BE-LIEVABLE! I am sooooo done with your constant lies. GET. OUT!"
        }
    ],
    "audio": {
        "format": "pcm16",
        "voice": "data:{MIME_TYPE};base64,$BASE64_AUDIO"
    },
    "stream": true
}'
```

  </TabItem>
</Tab>

## Price

- Billing: Free for a limited time.

- View Bill: You can view your usage on the [Billing](https://platform.xiaomimimo.com/#/console/usage) page in the Console.


--- DOCUMENT: Refer & earn ---
URL: https://mimo.mi.com/static/docs/quick-start/promotions/refer.md

# Refer & earn

## 1. Overview

{/* feishu-style:text-align:left */}
The Xiaomi MiMo Open Platform Refer & Earn program is a long-term referral activity. Any registered user can share a 6-character invite code or invite link with friends. Once a friend redeems the code, a referral relationship is created. When the friend completes their first paid Token Plan order, the inviter earns an additional 10% of the paid amount in credits, and the friend gets 10% off on the first order.

## 2. Eligibility

#### 2.1 Inviter

{/* feishu-style:text-align:left */}
Any registered Xiaomi MiMo Open Platform user is eligible to invite.

#### 2.2 Invitee

{/* feishu-style:text-align:left */}
Users who signed up to the Xiaomi MiMo Open Platform within the last 3 days. The following accounts are not eligible as invitees:

- Accounts older than 3 days

- Accounts that have already acted as an inviter or invitee

- Mutual invites (if A invited B, B cannot redeem A's code in return)

- Your own other accounts

## 3. Reward rules

#### 3.1 Credits back on paid orders

- **Trigger:**  Friend completes their first paid order (definition in §3.2)

- **Reward:** 

   - Inviter earns an additional 10% of the friend's paid amount in credits

   - Friend gets 10% off on the first order (stacks with any existing first-order discount)

   - Special bonus: Standard/Pro annual and Max monthly/annual plans receive a 20% referral bonus with no 30-transaction cap.

- **Crediting:**  To make sure rewards are paid out correctly, credits are released 3 days after your friend's first paid order (if the friend refunds within 3 days, the credits won't be paid out)

#### 3.2 First-paid-order definition

- **First order:**  The invitee's first paid Token Plan subscription within the lifetime of their account (does not include: API top-ups)

- **Paid amount:**  Actual cash payment after discounts (i.e. the amount charged to bank card / third-party payment)

- **30-day first-order window:**  The friend must complete the first paid order within 30 days of binding the invite code; otherwise the 10% paid-order reward will not be issued. The 10% first-order discount follows the same 30-day window: first orders completed after 30 days no longer get the discount.

- **Free Token Plan exclusion:**  If the friend has already received 1 month or more of free Token Plan, their first order no longer gets 10% off, and the purchase does not count toward referral rewards (no credits back will be issued)

## 4. Limits and quotas

{/* feishu-style:text-align:left */}
To prevent abuse,  the following two limits apply per account:

#### 4.1 Bindings: unlimited

- No friend limit: inviter can invite any number of friends to bind

- After 30 bindings, new invitees still get 10% off on Token Plan first order

- 10% off has no cap

#### 4.2 Credits-back quota: 30 paid orders

- First 30 paid orders earn 10% credits back; beyond 30, no further credits back, but friends still get 10% off

## 5. Using your rewards

#### 5.1 Credits redemption scope

{/* feishu-style:text-align:left */}
Credits can offset your Xiaomi MiMo model API call charges.

#### 5.2 Not redeemable for

- Token Plan subscription purchase

#### 5.3 Credit validity

- **Valid for 40 days:**  Counted from the credited date; unused balance after 40 days expires automatically

- **Deduction order:**  Earliest-expiring first (FIFO) · credits before cash

- Non-cashable, non-transferable, no change given

#### 5.4 Refund handling

- **Credits back (refund within 3-day pending window):**  The credits will not be issued

- **Credits back (refund after credited):**  The unused portion of the inviter's credits is immediately voided

- **Refunds do not affect counts:**  The 30-paid-order cap is counted by “paid order completed,” and counts are not reversed by refunds

## 6. Violations

{/* feishu-style:text-align:left */}
To keep things fair, the following will result in rewards being withheld or revoked:

- Inviting your own other accounts

- Mass registration via bots, virtual numbers, or other technical means

- Misleading promotion or fraudulent referrals

- Any other violation of these rules

> **About account identity:**  Xiaomi MiMo identifies accounts belonging to the same user using account bindings, login environment, network signals, and other signals.

## 7. Adjustments and sunset

{/* feishu-style:text-align:left */}
**This is a long-term feature with no fixed end date.** Xiaomi MiMo may adjust, pause, or sunset this feature based on operational needs.

## 8. Contact us

{/* feishu-style:text-align:left */}
For support or business inquiries:

- Email support-mimo@xiaomi.com

## 9. Final interpretation

{/* feishu-style:text-align:left */}
Within the limits of applicable law, Xiaomi MiMo holds final interpretation rights for these rules.

{/* feishu-style:text-align:left */}
Update Time August 13, 2026


--- DOCUMENT: Account and Authentication ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/account.md

# Account and Authentication

### Domestic/Overseas

- **What are the differences between overseas users and domestic users?** 

{/* feishu-style:text-align:left */}
Main differences:

{/* feishu-style:text-align:left */}
1) Different payment methods (domestic supports WeChat Pay, Alipay, and Xiaomi Pay; overseas payments are made through the Waffo Console (settled in US dollars $)).

{/* feishu-style:text-align:left */}
2) Return different Base URLs + Keys based on the region where the account is located, which are not interoperable.

{/* feishu-style:text-align:left */}
3) The overseas version currently has no invoicing function.

- **Can domestic and overseas usage be aggregated for calculation?** 

{/* feishu-style:text-align:left */}
No, it is not allowed. Different Base URLs and Keys are returned based on the region where the account is located, and usage is calculated separately. 

{/* feishu-style:text-align:left */}
<br></br>

### What login methods are supported? 

{/* feishu-style:text-align:left */}
The platform uses a Xiaomi account to log in. If you have already registered a Xiaomi account, you can log in directly; if you do not have a Xiaomi account, you can visit the Console to register, or register in advance at [id.mi.com](https://id.mi.com/): 

- Registration Methods for Mainland Chinese Users

   - Mobile Number Registration 

   - Third-party account authorized registration (supports registration via WeChat, Weibo, QQ, Alipay, and Apple ID; to ensure account security, a third-party account may be used to bind a commonly used mobile phone number)

- Registration Methods for Overseas Users

   - Email Registration

   - Third-party account authorized registration (supports Google/Facebook registration; to ensure account security, the email bound to the third-party account may be used as the security email for the Xiaomi account)

{/* feishu-style:text-align:left */}
<br></br>

### Unable to log in to the account? 

{/* feishu-style:text-align:left */}
If you have trouble logging in, please visit [ Xiaomi Account Help Center ](https://account.xiaomi.com/helpcenter?_locale=zh_CN) to view more account issues and solutions. 

{/* feishu-style:text-align:left */}
<br></br>

### Does recharge/purchase of a package require real-name authentication? 

{/* feishu-style:text-align:left */}
Domestic users need to complete personal or corporate real-name authentication before they can recharge or purchase a package. Both types of authentication can be used for purchase. 

{/* feishu-style:text-align:left */}
<br></br>

### What are the differences between personal certification and enterprise certification?

{/* feishu-style:text-align:left */}
Real-name authentication is divided into personal authentication and enterprise authentication, and the same account only supports authentication as one type. After completing real-name authentication, a personal account can be changed to an enterprise account through enterprise authentication, while an enterprise account cannot be changed to a personal account. 

{/* feishu-style:text-align:left */}
**The payment process for enterprise accounts is exactly the same as that for personal accounts,**  and enterprise accounts also support WeChat/Alipay/Xiaomi Pay. 

{/* feishu-style:text-align:left */}
<br></br>

### Why does it prompt that "the account has abnormal usage behavior" and cannot continue to be used?

{/* feishu-style:text-align:left */}
It may be because the system captured certain keywords during detection, triggering the automatic protection mechanism. We highly value the experience of each user and will conduct regular manual reviews to minimize the occurrence of misjudgments. If you encounter this issue, you can provide your user ID, and we will verify it for you and assist with the unblocking as soon as possible. 

{/* feishu-style:text-align:left */}
<br></br>

### Can I still continue to use the model service after canceling my Xiaomi account? 

{/* feishu-style:text-align:left */}
After the Xiaomi account is cancelled, the platform data will be cleared, and the API Key will expire synchronously. Please carefully evaluate and operate with caution. If you have any questions, please contact us  support-mimo@xiaomi.com  .


--- DOCUMENT: Payment ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/payment.md

# Payment

### How to recharge on the Open Platform? 

- **Pay-as-you-go API:**  Go to the [ Account Balance ](https://platform.xiaomimimo.com/#/console/balance) page to top up. The platform provides three top-up methods for domestic users: Xiaomi Pay, Alipay, and WeChat Pay; for overseas users, it provides  **commonly used top-up methods such as Apple Pay, Google Pay, credit/debit cards, etc.**  Top-ups generally arrive in real-time, and you can check the balance in the [ Account Balance ](https://platform.xiaomimimo.com/#/console/balance), and view the cumulative top-up amount and top-up records on the [ Top-up Details ](https://platform.xiaomimimo.com/#/console/recharge)  page. 

- **Token Plan:**  The plan does not currently support deduction using account balance or bonus, and you need to go to [Subscribe to Token Plan](https://platform.xiaomimimo.com/token-plan) to purchase separately.

{/* feishu-style:text-align:left */}
<br></br>

### Does purchasing a Token Plan package count towards cumulative recharge? 

{/* feishu-style:text-align:left */}
Subscriptions are not counted, and orders for subscription packages are not included in cumulative recharge. 

{/* feishu-style:text-align:left */}
<br></br>

### What payment methods are supported? 

{/* feishu-style:text-align:left */}
Domestically in China, it supports WeChat Pay, Alipay, and Xiaomi Pay, while overseas, it uses the Waffo payment gateway for payment (settled in US dollars $).

{/* feishu-style:text-align:left */}
<br></br>

### Is there a time limit for payment? 

{/* feishu-style:text-align:left */}
The effective payment duration is subject to the display on the page. After the timeout, the order will automatically close, and you need to place a new order. 

{/* feishu-style:text-align:left */}
<br></br>

### How to set up balance alerts?

{/* feishu-style:text-align:left */}
On the [Account Balance](https://platform.xiaomimimo.com/#/console/balance) page, you can enable the balance alert setting. Once enabled, when the account balance falls below the alert threshold, we will send a notification to your registered mobile number/email. Please check it carefully.

{/* feishu-style:text-align:left */}
<br></br>

### Does it support refund applications?

- **Pay-as-you-go API:**  Account balance supports full refund. If you have a refund request, you can open the contact pop-up window via the "Apply for Refund" button in the upper right corner of [Recharge Details](https://platform.xiaomimimo.com/#/console/recharge) , select the "Refund" option, and state the reason to initiate a refund request. The account balance will be refunded back to the original payment method after the review is approved (the consumed amount, the invoiced amount, and the platform-gifted amount cannot be refunded). After the refund request is accepted, you will no longer be able to continue calling the model service and will not be able to continue recharging. Billing may be delayed, and the refund amount shall be subject to the actual amount received. Refunds are generally processed and returned to the original payment method within 3-5 working days.

- **Token Plan:** Once a package is paid, it cannot be refunded.

{/* feishu-style:text-align:left */}
<br></br>

### How to issue an invoice?

- **Chinese User:** 

{/* feishu-style:text-align:left */}
Visit the [Invoice](https://platform.xiaomimimo.com/#/console/invoice) page, select the successfully recharged order, and issue an electronic invoice. Both individual and corporate invoices can be issued. Fill in your email or mobile phone number, and the invoice will be sent via email/sms upon completion of issuance.

{/* feishu-style:text-align:left */}
Note:
- The amount eligible for invoicing is the actual payment amount. Platform coupons, discounts, or amounts gifted by the platform cannot be invoiced. Refunded amounts cannot be invoiced. 
- According to regulations, invoices issued in the name of individuals can only be issued as digital electronic ordinary invoices, while invoices issued in the name of enterprises can be issued as digital electronic ordinary invoices or digital electronic special invoices.
- We generally issue invoices within 48 hours after receiving the application, and delays may occur in case of special circumstances. 
- Invoices support red stamping. If the invoice has been deducted or recorded, you need to log in to the Electronic Tax Bureau and confirm the information (red stamping confirmation form) within 72 hours to successfully complete the red stamping. After red stamping, the original order can be reissued with an invoice. 
- The invoicing entity is: Beijing Xiaomi Mobile Software Co., Ltd. 

- **Overseas Users:** 

{/* feishu-style:text-align:left */}
Each recharge order will automatically generate an invoice. When you complete a recharge, you can view the invoice on the order page. You can also enter the [Recharge Details](https://platform.xiaomimimo.com/#/console/recharge)  page to download historical invoices. 

{/* feishu-style:text-align:left */}
<br></br>

### Can I still call the API if my balance is insufficient? 

{/* feishu-style:text-align:left */}
Before the billing system goes live, models can be called for inference services normally when the balance is 0. 

{/* feishu-style:text-align:left */}
After the billing system goes live, due to a certain time delay, the balance may be ≤ 0. Once the balance becomes negative, the model inference service can no longer be used, and the next recharge order will first deduct the overdue amount. 

{/* feishu-style:text-align:left */}
<br></br>

###  Will there still be charges after the API Key is deleted?

- **Real-time inference API**: If the API Key is deleted, it will no longer be able to call the interface, and no charges will be incurred. The historical consumption records of this API Key can still be queried in [Billing](https://platform.xiaomimimo.com/#/console/usage). 

- **Batch inference API**: If the API Key is deleted, batch tasks already in execution will not be interrupted, and the billing will still be attributed to this API Key. The historical billing records of this API Key can still be viewed in  [Billing](https://platform.xiaomimimo.com/#/console/usage).


--- DOCUMENT: API Integration ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/api-integration.md

# API Integration

### How to obtain API Key? 

- **API Key for Pay-as-you-go API Calls:**  After logging in to the Xiaomi MiMo API Open Platform, apply for an API Key on the [Console - API Keys](https://platform.xiaomimimo.com/#/console/api-keys) page. When using the model via API, please include your API Key in the request header:` api-key: $MIMO_API_KEY ` or ` Authorization: Bearer $MIMO_API_KEY `. 

- **API Key of Token Plan:** After successful purchase, you can see the exclusive API Key on the [Token Plan ](https://platform.xiaomimimo.com/#/console/plan-manage)page. **Note: API Key is only visible and can be copied when created, please save it properly.** 

- **The API Key format for Token Plan is** `tp-xxxxx`**, which is only used for Token Plan subscription services; the API Key format for pay-as-you-go API calls is** `sk-xxxxx`**, used for pay-as-you-go billing. The two are independent of each other and cannot be mixed. The API Key for Token Plan is only available within the validity period of the Token Plan package you have subscribed to.** 

{/* feishu-style:text-align:left */}
<br></br>

### What if the API Key is lost or leaked? 

{/* feishu-style:text-align:left */}
can be reset on the [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) page. 

{/* feishu-style:text-align:left */}
<br></br>

### How can I obtain the Base URL of the Token Plan? 

{/* feishu-style:text-align:left */}
The Base URL provided on the Token Plan page shall prevail: Two types of Base URLs are provided, one compatible with the OpenAI interface protocol and the other compatible with the Anthropic interface protocol, which can be copied and used as needed. 

{/* feishu-style:text-align:left */}
<br></br>

### Which programming tools does Token Plan support? 

{/* feishu-style:text-align:left */}
Supports mainstream programming tools and model frameworks, such as Claude Code, OpenClaw, OpenCode, Kilo Code, Cline, Hermes Agent, CodeBuddy Code, etc. For specific access methods, please refer to [Overview of AI Tools](https://platform.xiaomimimo.com/#/docs/integration/tools-overview).

{/* feishu-style:text-align:left */}
<br></br>

### Can Token Plan be used in multiple programming tools at the same time? 

{/* feishu-style:text-align:left */}
The same package can be used across all supported tools, but the quota is shared, and usage of all tools will consume the same package quota. 

{/* feishu-style:text-align:left */}
<br></br>

### What's the difference between OpenAI and Anthropic interfaces?

- OpenAI interface `/v1/chat/completions`  follows OpenAI format, including developer/system/user/assistant roles

- Anthropic interface  `/anthropic/v1/messages` follows Claude format, with a separate system parameter

{/* feishu-style:text-align:left */}
<br></br>

### How to make multi-turn tool calls in thinking mode?

{/* feishu-style:text-align:left */}
During the multi-turn tool calls process in thinking mode, the model returns a `reasoning_content` field alongside `tool_calls`. To continue the conversation, it is recommended to keep all previous `reasoning_content` in the `messages` array for each subsequent request to achieve the best performance.

{/* feishu-style:text-align:left */}
The requested example is as follows:

```bash
curl --location --request POST 'https://api.xiaomimimo.com/v1/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "messages": [
        {
            "role": "assistant",
            "content": "Hello! I am MiMo.",
            "reasoning_content": "Okay, the user just asked me to introduce myself. That is a pretty straightforward request, but I should think about why they are asking this."
        },
        {
            "role": "user",
            "content": "What is the weather like in Hebei?"
        }
    ],
    "model": "mimo-v2.6-pro",
    "max_completion_tokens": 1024,
    "temperature": 1.0,
    "stream": false,
    "tools": [
        {
            "type": "function",
            "function": {
                "name": "get_current_weather",
                "description": "Get the current weather in a given location",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "location": {
                            "type": "string",
                            "description": "The city and state, e.g. San Francisco, CA"
                        },
                        "unit": {
                            "type": "string",
                            "enum": [
                                "celsius",
                                "fahrenheit"
                            ]
                        }
                    },
                    "required": [
                        "location"
                    ]
                }
            }
        }
    ],
    "tool_choice": "auto"
}'
```

{/* feishu-style:text-align:left */}
<br></br>

### Why aretool_calls sometimes included in the reasoning_content field and sometimes in a separate tool_calls field?

{/* feishu-style:text-align:left */}
The appearance of `tool_calls` in the reasoning content indicates instability and incomplete output caused by the model having `thinking` enabled when calling `tool`. It is recommended to disable `thinking` when calling `tool` calls and to adjust the settings according to [Model Hyperparameters](https://platform.xiaomimimo.com/#/docs/quick-start/model-hyperparameters) to achieve a more stable and better user experience.

{/* feishu-style:text-align:left */}
<br></br>

### What's the response speed?

{/* feishu-style:text-align:left */}
Response speed depends on:

- Request length and complexity

- Server load and geographic location

- Whether streaming response is used

{/* feishu-style:text-align:left */}
<br></br>

### How to handle timeouts?

{/* feishu-style:text-align:left */}
Please implement reasonable timeout handling on the client side:

- Set reasonable connection and read timeout times

- Use exponential backoff for retries

- For long responses, it's recommended to use streaming mode

{/* feishu-style:text-align:left */}
<br></br>

### What if the API returns inappropriate content?

{/* feishu-style:text-align:left */}
The platform has added content review for both user input and model output. If violations occur, the returned content will be automatically intercepted to ensure the content you receive is safe.

{/* feishu-style:text-align:left */}
<br></br>

### Why doesn’t the model perform a web search after enabling online search? 

{/* feishu-style:text-align:left */}
There may be three reasons:

- **Cache**: There is a 5-minute cache period after enabling / disabling online search. The online search switch will not take effect immediately within 5 minutes.

- **Model determines no need for search**: The model judges that the current query does not involve real-time information and can be answered directly with its own knowledge. To force a search, set `forced_search: true`.

- **Only some models are supported**: Currently only `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro` and `mimo-v2.5` supports web search.

{/* feishu-style:text-align:left */}
<br></br>

### Does it support local file upload?

- Real-time Inference, local file upload is not supported at this time.

- Batch Inference, supporting the upload of local files (in JSONL format).


--- DOCUMENT: Plans and Pricing ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/token-plan/Plans&Pricing.md

# Plans and Pricing

- **What packages are available for the Token Plan?** 

{/* feishu-style:text-align:left */}
We currently offer four subscription plans: Lite (available for individual users only), Standard, Pro, and Max. Both monthly and annual auto-renewal subscriptions are supported, while team plans additionally support one-time purchases for custom durations (1/3/6/12 months). Each plan comes with distinct usage quotas and benefits.

- **Which models does Token Plan support?** 

{/* feishu-style:text-align:left */}
Both the individual and team versions support a total of **6** **models, including the MiMo-V2.6 series**, which are accessible across all subscription tiers.

> mimo-v2.6-pro: Flagship Full-Modal Inference Model
>
> mimo-v2.6-flash: An Ultra-Full-Modal Inference Model with Extreme Speed
>
> mimo-v2.5-asr: Speech Recognition Model
>
> mimo-v2.5-tts/ v2.5-tts-voiceclone / v2.5-tts-voicedesign: Speech synthesis model (limited-time free)

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**The Individual edition additionally supports mimo-v2.5-pro and mimo-v2.5.Both of these two models will be officially taken offline at 10:00 on October 21, 2026 Beijing Time, and it is recommended to switch to the new version of the models as soon as possible.**

</div>

- **How much is the package deal? Are there any discounts available?** 

{/* feishu-style:text-align:left */}
The specific prices of the four packages are subject to those displayed on the landing page. The platform currently offers the following time-limited promotional activities:
- First-purchase Discount: Users purchasing the Individual Edition for the first time can enjoy a 12% discount, with only one discount eligible per account; the Team Edition is not eligible for this discount.
- Annual auto-renewal plan: Enjoy a 12% discount compared to the monthly auto-renewal plan; first-purchase offers are not applicable to the annual plan.
- Night discount rate: During off-peak hours (Beijing time 00:00–08:00, i. e. UTC 16:00–24:00), the Credits consumption coefficient is 0.8x.

- **Is continuous subscription supported?** 

{/* feishu-style:text-align:left */}
Supported. Two subscription cycles are available: monthly auto-renewal and annual auto-renewal, with automatic renewal upon expiration. You may cancel the auto-renewal at any time on the Token Plan page. The annual auto-renewal plan offers a 12% discount, which is a more favorable deal compared to the monthly auto-renewal plan.

- **Can I purchase multiple packages or upgrade my package?** 

{/* feishu-style:text-align:left */}
**You can purchase both the Individual Plan and the Team Plan at the same time, but only one subscription per plan type can be active simultaneously.** 
- **Individual plan: If you wish to get more credits before your current plan expires**, you can convert the Credit quota you have already used into the equivalent amount, then pay the price difference on that basis to upgrade to a higher plan and get more Credits. Cross-tier plan upgrade by paying the price difference is supported, while plan downgrade is not allowed. If you have already upgraded to the top-tier Max plan, further upgrade is unavailable. **After your current plan expires, you can repurchase a plan of any tier.** 
> Price difference = New package price -(Remaining balance of original package / Total amount of original package) * Original package price

- **Team Plan:** When the monthly quota for a seat is exhausted, that seat will be suspended from service, without affecting other seats that still have remaining quota. If you need to continue using the service, you can contact the team owner to purchase additional seats as needed, or switch to the pay-as-you-go standard API service. Plan upgrades and downgrades are temporarily not supported. A team-shared usage package is coming soon, stay tuned.


--- DOCUMENT: Validity and Expiry ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/token-plan/Validity&Expiry.md

# Validity and Expiry

- **How long is the package valid after purchase?** 

{/* feishu-style:text-align:left */}
It takes effect immediately upon purchase, and is valid for one calendar month/year from the date of purchase based on UTC time.

{/* feishu-style:text-align:left */}
For example, if you subscribe to a certain monthly plan on March 28, the plan will expire at 23:59:59 UTC on April 28.

- **Will the subscription be automatically renewed after it expires?** 

{/* feishu-style:text-align:left */}
It depends on whether you have enabled auto-renewal. If auto-renewal is enabled, upon the expiration of your plan, the system will initiate automatic deduction via Alipay/WeChat/Mi Pay (for domestic users) or via waffo/Stripe (for overseas users). After successful deduction, your subscription will automatically enter the next billing cycle without any manual operation required; if auto-renewal is not enabled, the service will be suspended once your plan expires, and you will need to manually resubscribe.

- **My credit limit has been used up but it hasn't expired yet. Can I still use it?** 

{/* feishu-style:text-align:left */}
No.

{/* feishu-style:text-align:left */}
**Individual Edition**: When either "expiration" or "all Credits limit is used up" is met, the service will be stopped. The system will not continue to consume your bonus or account balance. We support package upgrade function. Regardless of your package consumption, we support automatically converting your remaining Credit into an equivalent amount. You can upgrade the package by making up the difference to obtain more Credits. If you need to continue using it, please upgrade the package or switch to the pay-as-you-go API.

{/* feishu-style:text-align:left */}
**Team Edition:**  After a single seat quota is exhausted, the seat will not be available for the month. But it does not affect the use of other seats that have not been exhausted. You can wait for quota reset, add purchase and reassign seats, or switch to normal API.

- **My subscription plan has expired before I've used up its benefits. Can the remaining balance be rolled over?** 

{/* feishu-style:text-align:left */}
No. The service will be terminated immediately upon the expiration of your current plan. You will need to subscribe to a new plan separately, and any unused quota from your old plan will not be carried over to the new one.

- **Will I receive a reminder when my data plan expires or is about to run out?** 

{/* feishu-style:text-align:left */}
Yes. If auto-renewal is enabled for your plan, you will receive renewal reminders via SMS, email, and in-app messages on payment apps (WeChat/Alipay) five days before the expiration date; if auto-renewal is not enabled for your plan, you will receive reminders via SMS and email two days before the expiration date and on the expiration date itself.

- **Will I get a reminder when my package quota is almost used up?** 

{/* feishu-style:text-align:left */}
Yes. You will receive SMS and email alerts when your current plan usage reaches 50%, 90% and 100% respectively.


--- DOCUMENT: Usage and Quota ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/token-plan/Usage&Quota.md

# Usage and Quota

- **Are the quotas of different models consumed independently?** 

{/* feishu-style:text-align:left */}
The available models in the package are consumed in parallel at different ratios, rather than independently. All TTS models are free for a limited time and do not consume package credits. ASR models deduct credits based on the duration of the input audio (the duration is counted to the nearest second and finally converted to hours for statistics). The table below lists the credit deduction per token for **Cache, input and output** of each model.

{/* feishu-style:text-align:left */}
Language Model Language Model

<table>
<colgroup>
<col style="width: 19.37%" />
<col style="width: 28.49%" />
<col style="width: 31.2%" />
<col style="width: 20.939999999999998%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input (Cache Hit) Token</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input (cache miss) Token</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Output Token</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.6-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2.5 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">300 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">600 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.6-flash</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">200 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2.5 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">300 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">600 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">200 Credits</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
ASR Model

<table>
<colgroup>
<col style="width: 40.48%" />
<col style="width: 59.52%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input audio duration (h)</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-asr</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">30M Credits</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
The TTS series models are available for free for a limited time and will not consume your package credits.

- **What is the off-peak 0.8x coefficient**

{/* feishu-style:text-align:left */}
To balance resource pressure and pass on benefits to users, the Credit consumption coefficient will be 0.8 times when using the model during off-peak hours (Beijing time 00:00-08:00, i. e. UTC 16:00-24:00).

{/* feishu-style:text-align:left */}
For example, in a scenario where you use the mimo-v2.5-pro model and consume 10M Credits during peak hours, only 8M Credits will be consumed during off-peak hours.

- **What should I do if I'm worried that Credits will be consumed quickly since the usage is calculated based on Tokens?** 

   - We recommend that you check your historical token usage in each AI Agent framework before making a purchase, and select an appropriate plan based on your actual usage.

   - We have designed a Progress Bar system, and you can view the progress and make advance planning in the Token Plan.

   - **For the Individual plan**, we offer a package upgrade feature. Regardless of your current package usage, we will automatically convert your remaining Credits into an equivalent monetary value, allowing you to upgrade your plan by paying the price difference to get more Credits. **For the Team plan**, you can add seats mid-subscription. Meanwhile, we are also developing the shared usage package feature, so stay tuned.

- **What is the remaining value of a service package?** 

{/* feishu-style:text-align:left */}
When you renew or upgrade your current unused Individual Plan, the system will calculate the equivalent value based on the Credits consumption of the current plan, and this remaining value will be used to offset part of the payment for the new plan. Please note that for users who participated in the platform's Credits giveaway activities before August 25,2026, and each subsequent package payment amount is less than ￥7 (domestic China) / 1$(overseas), the remaining value will be calculated based on your last actual payment, which can be viewed in the renewal/upgrade pop-up window for details.

- **Why are there still compensation Credits on my Token Plan page?** 

{/* feishu-style:text-align:left */}
During the process of renewing your current plan, since the remaining value of your previous plan is higher than the value of the current plan, the platform will compensate you with Credits equivalent to the value of the price difference.


--- DOCUMENT: Team Edition Benefits Related ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/token-plan/Team-Benefits.md

# Team Edition Benefits Related

- **What is the difference between the Token Plan Individual Edition and Team Edition?** 

{/* feishu-style:text-align:left */}
The Team version supports seat-based purchase, unified payment, and centralized management of members and usage, making it ideal for teams or enterprises requiring multi-person collaboration. It maintains the same pricing as the Individual version and inherits all the benefits of the Individual version(with the exception of the two models mimo-v2.5-pro and mimo-v2.5).

- **How to invite members? Is bulk invitation supported?** 

{/* feishu-style:text-align:left */}
Members shall first register a Xiaomi account, and provide the account owner with their full name, Xiaomi account ID, as well as a mobile phone number or email address for receiving notifications. At least one of the mobile phone number and email address must be filled in; if both are provided, notifications will be sent via both SMS and email simultaneously. The owner may invite members manually, or download the template, fill in the required information and then send bulk invitations.

- **Why didn't the member receive the notification after I sent the invitation?** 

{/* feishu-style:text-align:left */}
Please confirm that the mobile phone number or email address you filled in when receiving the invitation is correct and can normally receive notifications, and check the SMS interception records or the junk email folder. If you still have not received it, the owner can copy the unexpired invitation link of this member and directly provide it to the member.

- **Why can't I join the team after opening the invitation link?** 

{/* feishu-style:text-align:left */}
Invitation is valid for 7 days from the date of sending. The invitation link will become invalid if the invitation expires, the member is removed by the owner, or the team subscription expires. Please check whether the Xiaomi ID of the logged-in account matches the invited ID, and whether the account's region matches the team.

- **What is the difference between an owner and an ordinary member? Does the owner need a seat for Individual use?** 

{/* feishu-style:text-align:left */}
The owner is responsible for purchasing plans, inviting and managing members, assigning or revoking seats, and can view the team's overall usage, member usage and transaction information. Regular members can only view their own seats, call information and usage details.

{/* feishu-style:text-align:left */}
When the owner invokes the model, seats also need to be allocated, and the quota of the allocated seats will be consumed.

- **I have already joined the team, but why is the API Key still unavailable?** 

{/* feishu-style:text-align:left */}
Once a seat is revoked, a member is removed, or the plan expires, the Team Edition API Key will no longer be valid. Please ensure that you have accessed the correct team workspace, been assigned a seat with remaining quota, and that the plan has not expired, then use your exclusive API Key for this workspace and the Base URL corresponding to the applicable protocol.

- **What should I do if the seat usage is insufficient? Will the quota exhaustion affect other members?** 

{/* feishu-style:text-align:left */}
Team packages are billed independently per seat. Once the monthly quota for a given seat is exhausted, that seat will be suspended from service, while other seats with remaining quota will remain unaffected.

{/* feishu-style:text-align:left */}
If you wish to continue using the service, you may switch to the pay-as-you-go standard API service. **The team-shared quota package feature is currently under development, stay tuned**.

- **If I purchase an annual plan or buy multiple months at one time, will I get all the quota at once?** 

{/* feishu-style:text-align:left */}
No. Whether it is monthly auto-renewal, annual auto-renewal, or one-time purchase of multiple months for the Team Plan, the seat quota will be reset on a monthly basis.

{/* feishu-style:text-align:left */}
Standard, Pro, and Max tiers offer per seat per month respectively **11 billion, 38 billion, and 82 billion Credits**. Please check the specific reset time on the seat page.

- **Can the unused quota this month be carried over to next month?** 

{/* feishu-style:text-align:left */}
No. The quota of Team Edition seats is reset on a monthly basis, and any unused quota in the current month will not be carried over to the next month. This rule also applies to annual continuous subscription plans and multi-month one-time purchase plans. The specific reset time is subject to the display on the seat page.

- **After a member leaves, can the seat be allocated to another person? Will the quota be recalculated?** 

{/* feishu-style:text-align:left */}
Yes. The owner may reclaim a member seat, or assign a seat that automatically reverts upon member removal to another member.

{/* feishu-style:text-align:left */}
When reallocating a consumed seat, the new member inherits the remaining quota of that seat for the current billing cycle, the consumed portion will not be reset, and historical consumption will be retained in the seat usage record.

- **How are additional seats purchased mid-subscription charged? When will the new seats expire?** 

{/* feishu-style:text-align:left */}
The owner may purchase additional seats of the same tier during the validity period of the current plan, with a minimum purchase of 1 seat per transaction. The amount for additional purchase shall be calculated according to the following formula:

> **Additional purchase amount = (Remaining billing duration ÷ Total billing duration of the current period) × Price per seat billing cycle × Quantity of additional purchases**

{/* feishu-style:text-align:left */}
The quota for the first usage cycle of the newly added seats shall also be converted based on the remaining validity period, and the expiration time shall be consistent with that of the current package. After the additional purchase is successful, the owner can assign the seats to members for use, and the specific amount and quota shall be subject to the display on the additional purchase page.

- **Is it possible to upgrade or downgrade the package, or unsubscribe from some of the seats?** 

{/* feishu-style:text-align:left */}
During the validity period of the current plan, changing the plan tier or unsubscribing from individual or partial seats is temporarily unavailable. If you need additional usage resources, you may purchase extra seats for the current tier as needed. Once purchased, the subscription service takes effect immediately; no refunds are supported, and unused quotas are non-refundable.

- **Can I turn off auto-renewal? What will happen after the subscription expires?** 

{/* feishu-style:text-align:left */}
Monthly and annual auto-renewal subscriptions support automatic renewal, which can be turned off by the subscriber at any time.

{/* feishu-style:text-align:left */}
After the plan expires, team seats will become unassigned, the Team Edition API Key will no longer be usable, and pending invitation links will also become invalid; historical usage details will still be retained. Owners can repurchase via the "Subscribe Now" entry in the plan overview.

- **How to make payment for a large amount?** 

{/* feishu-style:text-align:left */}
There are certain payment limits for monthly and annual auto-renewal subscriptions, subject to the display on the relevant page. If you receive a limit-exceeded prompt when attempting to sign up for auto-renewal, it is recommended to switch to one-time purchase. The actual payment shall still be subject to the limits set by your selected payment channel and issuing bank.

- **I already have an individual plan. Can I still use the team plan? Is it possible to join multiple teams?** 

{/* feishu-style:text-align:left */}
Yes. Individual plan and team plan can take effect at the same time, using their own API Key and quota respectively. You can also join multiple teams and use the seat rights allocated to the corresponding team; When calling, please configure the exclusive API Key in the corresponding team space.


--- DOCUMENT: Token Plan and Desktop Membership ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/token-plan/desktop-guide.md

# Token Plan and Desktop Membership

## Overview

{/* feishu-style:text-align:left */}
Token Plan and Xiaomi MiMo Desktop Membership are two independent subscription services that differ in terms of application scope, supported models, benefits and usage quotas:

- **Token Plan：** A subscription plan that allows users to access the MiMo model via a dedicated API Key on Desktop and other supported AI tools, available in both individual and team editions.

- **Desktop Membership:** A membership service exclusive to the Desktop version, covering the usage quotas and benefits of the corresponding plan. It is suitable for users who directly use the MiMo large language model on the desktop to complete various tasks.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**The benefits, pricing, and quotas of the two subscriptions are independent of each other.** Purchasing a Token Plan does not automatically activate a Desktop membership, nor does purchasing a Desktop membership entitle you to a Token Plan package.

**Users who have subscribed to the Token Plan can use the Desktop function normally after completing the access configuration, and there is no need to subscribe to an additional Desktop membership.** Token Plan package quota will be consumed during use, and the mimo-v2.6-pro-ultraspeed model is not supported for the time being.

</div>

## Rights and Scope of Application

<table>
<colgroup>
<col style="width: 14.42%" />
<col style="width: 55.53%" />
<col style="width: 30.049999999999997%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Comparison Item</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Xiaomi MiMo Desktop Membership</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Subscription Content</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Call the specified MiMo model services, package quotas and benefits in supported tools</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Member services, package quotas and benefits within Desktop</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Purchase Discounts</span></td>
<td><ul><li>First-purchase discount: Get 12% off your first Personal plan purchase, once per account. Team plans are not eligible.</li><li>Annual subscriptions: Both Personal and Team plans offer 12% off compared with monthly subscriptions. The first-purchase discount does not apply to annual subscriptions.</li><li>Off-peak usage rate: During off-peak hours (00:00–08:00 Beijing time, or 16:00–24:00 UTC), usage is charged at a 0.8× Credits consumption rate.</li></ul></td>
<td><ul><li>Annual subscriptions:  Offer 12% off compared with monthly subscriptions.</li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Subject to the actual display on the page.</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Scope of Application</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Compatible with supported AI tools including Desktop, OpenClaw, Claude Code, Codex, etc.</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Member benefits and quotas are only available for use within the Desktop app</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Usage in Desktop</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Configure a dedicated API Key for the Token Plan and the corresponding service address via custom model settings</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Log in to the subscribed membership Xiaomi account and use the corresponding membership services</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Desktop Features</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">All functions can be used normally after access</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">All function properly</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Model Equity</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Subject to the supported model list announced in the Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Subject to the model benefits announced for each membership tier on Desktop</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">UltraSpeed Model</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Not supported yet</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Currently included in the "Premium" and "Elite" membership tiers, but not included in the "Basic" and "Intermediate" membership tiers</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Credit Utilization</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Consumes Token Plan Credits, and multiple integrated tools share the package quota</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Consume the usage quota corresponding to the Desktop membership</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Whether it includes another subscription</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Excludes Desktop membership</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Excluding the Token Plan package</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
For details on supported plans and models, please refer to the [Token Plan Subscription Guide ](https://mimo.mi.com/docs/zh-CN/tokenplan/Token%20Plan/subscription)and [Desktop Membership Plan](https://mimo.xiaomimimo.com/pricing/).

## How to choose

- Token Plan

- In addition to Desktop, you want to use the MiMo model in other supported tools (such as Codex, ClaudeCode, OpenCode, OpenClaw)

- Desktop Membership

- Hope to use it directly in Desktop without configuring an API Key

- Hope to use UltraSpeed in Desktop

## Frequently Asked Questions

- **I have purchased the Token Plan, but why does the Desktop app still prompt that my membership has not been activated?** 

{/* feishu-style:text-align:left */}
The membership status in Desktop refers to **the Desktop membership subscription status**. The Token Plan is a standalone model subscription package, and purchasing it will not activate the Desktop membership.

{/* feishu-style:text-align:left */}
If you want to use the Token Plan in Desktop, please configure the exclusive credentials for the package via custom models. After integration, you can use Desktop functions normally without subscribing to an additional Desktop membership.

{/* feishu-style:text-align:left */}
For configuration instructions, please refer to [MiMo Desktop Access Guide](https://mimo.mi.com/docs/zh-CN/tokenplan/integration/mimo-desktop).

- **Are there any feature restrictions when accessing Desktop via the Token Plan?** 

{/* feishu-style:text-align:left */}
All features of Desktop are fully functional. The scope of supported models is subject to the model list specified in the Token Plan; mimo-v2.6-pro-ultraspeed is not temporarily supported. Usage will consume the quota of your Token Plan package, which is shared with other integrated tools.

- **Why can't I use UltraSpeed after purchasing the Token Plan? Can I use it by upgrading my Token Plan tier?** 

{/* feishu-style:text-align:left */}
None of the current Token Plan tiers include **mimo-v2.6-pro-ultraspeed**, and upgrading your Token Plan tier will not grant you access to this model. If you wish to use UltraSpeed via a Desktop membership, you currently need to subscribe to the "Premium" or "Elite" Desktop membership tier.

- **Can the Desktop membership quota be used for other AI tools?** 

{/* feishu-style:text-align:left */}
No. The benefits and quota of Desktop membership are only valid within Xiaomi MiMo Desktop, and do not include the quota of the Token Plan package for other tools.

{/* feishu-style:text-align:left */}
If you need to call the MiMo model in other supported AI tools, you can select the Token Plan and complete the configuration according to the access guide of the corresponding tool.


--- DOCUMENT: Promotions ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/promotions.md

# Promotions

### What promotions and benefits are currently available?

{/* feishu-style:text-align:left */}
The main benefit currently offered is bonus credit, which can be used to offset pay-as-you-go API usage and quickly try MiMo models. You can receive bonus credit through **Refer & Earn** or new user registration. Eligibility, credit amounts, and payout rules vary by campaign. Bonus credit is currently valid for 40 days and expires after that.

{/* feishu-style:text-align:left */}
<br></br>

### How do I join "Refer & Earn"?

{/* feishu-style:text-align:left */}
Any registered user can invite; new users (signed up within 3 days) can be invited. After your friend enters via your invite link and the code is bound, only the referral relationship is created. When your friend completes their first paid order, you earn an additional 10% of their paid amount in credits and they get 10% off.

{/* feishu-style:text-align:left */}
<br></br>

### Where is my invite code? How do I share it?

{/* feishu-style:text-align:left */}
After signing in, use the "Invite friends" feature to view your 6-character invite code. You can copy the invite link or save the invite poster to share with friends.

{/* feishu-style:text-align:left */}
<br></br>

### How does my friend redeem the code?

{/* feishu-style:text-align:left */}
After your friend clicks your invite link and signs in to the console, the system will automatically bind the invite code (no manual entry required). Once the binding succeeds, only the referral relationship is created.
If the invite code is not filled in automatically after opening the link, your friend can manually enter the 6-character code in the console to bind. Each invitee can redeem only 1 invite code. The binding cannot be changed afterward.

{/* feishu-style:text-align:left */}
<br></br>

### When will I earn credits back?

{/* feishu-style:text-align:left */}
Binding an invite code only creates the referral relationship. If your friend later completes a first paid order, Tier 2 (Credits back) kicks in — you earn 10% of their paid amount. The invitee receives an additional 10% off on top of the existing first-order offer. Eligible credits are issued 3 days after the order is completed.

{/* feishu-style:text-align:left */}
<br></br>

### What counts as a "first paid order"?

{/* feishu-style:text-align:left */}
A "first paid order" is your friend's first paid Token Plan subscription within the lifetime of their account.
Includes: Token Plan subscription first payment
Does not include: API top-ups; unpaid orders
30-day first-order window: Your friend must complete their first paid order within 30 days of binding the invite code; after 30 days, credits back will no longer be issued. The 10% first-order discount follows the same 30-day window: first orders completed after 30 days no longer get the discount.
Free Token Plan exclusion: If your friend has already received 1 month or more of free Token Plan, their first order no longer gets 10% off, and the purchase does not count toward referral rewards (no credits back will be issued).
A refund does not reset first-order eligibility.

{/* feishu-style:text-align:left */}
<br></br>

### Is there a cap on credits back?

{/* feishu-style:text-align:left */}
Credits-back quota: 30 paid orders — first 30 paid orders earn 10% credits back. Beyond 30 orders, the inviter no longer earns credits back, while invitees can still receive 10% off.
Special bonus: Standard annual, Pro annual, and Max (any duration) members earn 20% credits back on paid orders, with no 30-order cap.

{/* feishu-style:text-align:left */}
<br></br>

### What can credits be used for?

{/* feishu-style:text-align:left */}
Credits can only offset Xiaomi MiMo model API call charges.
Redeemable for: API call charges
Not redeemable for: Token Plan subscription
Non-cashable, non-transferable, no change given
Valid for 40 days: counted from the credited date, earliest-expiring first (FIFO); unused balance after 40 days expires automatically.

{/* feishu-style:text-align:left */}
<br></br>

### Does my friend's refund affect my rewards?

{/* feishu-style:text-align:left */}
Credits back (refund within 3-day pending window): the credits will not be issued
Credits back (refund after credited): the unused portion of the inviter's credits is immediately voided
Refunds do not affect counts: the 30-paid-order cap is counted by "paid order completed", and counts are not reversed by refunds

{/* feishu-style:text-align:left */}
<br></br>

### My friend redeemed my code, but I don't see the reward.

{/* feishu-style:text-align:left */}
Check in order:

1. Confirm the binding succeeded: a successful binding only creates the referral relationship.

1. Credits back arrive 3 days after the friend's first paid order: please wait 3 days

1. Check the 30-day first-order window: your friend must complete the first paid order within 30 days of binding; after 30 days, no credits back are issued

1. Check the cap: After 30 paid referrals, the 10% bonus stops — unless you are on Standard annual, Pro annual, or Max monthly/annual plan.

1. If the invitee has already received at least 1 month of free Token Plan, the first paid order does not receive 10% off or qualify for credits back.

{/* feishu-style:text-align:left */}
<br></br>

### What should I do if the invite code cannot be bound?

{/* feishu-style:text-align:left */}
Check whether the invite code is correct, the invitee is using their own code, the two accounts already have a reverse referral relationship, or the invitee registered within the last 3 days. If the page says the account is restricted, email support-mimo@xiaomi.com. If there is a network error or the page is unresponsive, try again later.


--- DOCUMENT: Others ---
URL: https://mimo.mi.com/static/docs/quick-start/faq/others.md

# Others

## Model Deprecation

{/* feishu-style:text-align:left */}
With the continuous iteration of the MiMo model, the new version has comprehensively outperformed the old version in terms of effectiveness and performance. We will gradually deprecate the legacy models, and the specific plan will be announced in advance via SMS, email, website announcements, etc. Please pay attention to the relevant messages and complete the switch in a timely manner.For information on models that have been or will be deprecated, please refer to [Model Deprecation](https://platform.xiaomimimo.com/docs/zh-CN/updates/deprecate).

## Model Capabilities and More Usage Methods

### How is the experience of the Xiaomi MiMo model?

1. Ordinary users can directly experience online through the official website **Xiaomi MiMO Studio (**  [**https://aistudio.xiaomimimo.com**](https://aistudio.xiaomimimo.com/) **)**  

1. Developers can obtain resources through Xiaomi's official channels:

   1. GitHub repository ( [https://github.com/xiaomimimo/MiMo-V2-Flash](https://github.com/xiaomimimo/MiMo-V2-Flash) ) to obtain open-source model weights and code

   2. Xiaomi MiMo Open Platform ( [https://platform.xiaomimimo.com/](https://platform.xiaomimimo.com/) ) Apply for API Service

1. It should be noted that Xiaomi has not yet launched an official MiMo standalone app. Downloading related software from unofficial channels poses security risks, so please do not download it.

{/* feishu-style:text-align:left */}
<br></br>

### Does the model support local deployment? How can I perform local deployment? 

{/* feishu-style:text-align:left */}
The large model released this time has been open-sourced, please see:  https://github.com/XiaomiMiMo/MiMo-V2-Flash. Local deployment primarily supports sglang, for related issues please refer to:  https://github.com/sgl-project/sglang/pull/15207. 

{/* feishu-style:text-align:left */}
<br></br>

### Why does the message "Server is busy, please try again later" always appear during the conversation?

{/* feishu-style:text-align:left */}
Currently, MiMo's output is affected by various factors such as server load, question content, and model output. You can start a new conversation or try regenerating to get a better communication experience.

{/* feishu-style:text-align:left */}
<br></br>

### When will the model support the function of uploading files? 

{/* feishu-style:text-align:left */}
Support will be provided as soon as possible, please wait patiently. 

## Contact Us

{/* feishu-style:text-align:left */}
If you need any assistance or business consultation, please feel free to contact us at any time. 

- Email sent to [support-mimo@xiaomi.com](mailto:support-mimo@xiaomi.com) 

- Scan the QR code in the lower left corner to join the developer communication group 

- Fill out [ Contact Us ](https://platform.xiaomimimo.com/#/contact) Questionnaire Feedback


--- DOCUMENT: Rate Limit ---
URL: https://mimo.mi.com/static/docs/api/guidance/rate-limit.md

# Rate Limit

{/* feishu-style:text-align:left */}
This page lists all the models currently supported by the Xiaomi MiMo API Open Platform and their rate-limiting quotas, helping you plan your request frequency before integration. 

### Rate Limiting Instructions

{/* feishu-style:text-align:left */}
The platform sets a model concurrency limit for each account. When the server load is high, response delays or ` 429 ` error may occur. We recommend that you plan your request frequency reasonably and implement request retry and backoff strategies in high-concurrency scenarios to avoid triggering rate limits. 

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

- **RPM (Requests Per Minute)** : The maximum number of requests initiated per minute. The calculation scope is the sum of the total number of requests from all API Keys under a single account when calling the same model.

- **TPM (Tokens Per Minute)** : The maximum number of Tokens that can be interacted with per minute. The calculation scope is the sum of the total number of requested Tokens for all API Keys under a single account when calling the same model.

</div>

### Text Generation Model

<table>
<colgroup>
<col style="width: 44.04%" />
<col style="width: 27.98%" />
<col style="width: 27.980000000000004%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Model ID</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">RPM</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">TPM</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">10M</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">10M</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro-ultraspeed`</span></td>
<td colspan="2"><span style="display: inline-block; width: 100%; text-align: left;">Customized services available, please [contact us](https://platform.xiaomimimo.com/contact?userId=2451887661)</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-pro`(to be deprecated)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">10M</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5`(to be deprecated)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">10M</span></td>
</tr>
</tbody>
</table>

### Automatic Speech Recognition Model (ASR) 

<table>
<colgroup>
<col style="width: 44.04%" />
<col style="width: 27.98%" />
<col style="width: 27.980000000000004%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Model ID</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">RPM</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">TPM</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-asr`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">10K</span></td>
</tr>
</tbody>
</table>

### Text-to-Speech (TTS) Model

<table>
<colgroup>
<col style="width: 44.04%" />
<col style="width: 27.98%" />
<col style="width: 27.980000000000004%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Model ID</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">RPM</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">TPM</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">10M</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts-voiceclone`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">10M</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-tts-voicedesign`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">10M</span></td>
</tr>
</tbody>
</table>


--- DOCUMENT: Model Hyperparameters ---
URL: https://mimo.mi.com/static/docs/api/guidance/model-hyperparameters.md

# Model Hyperparameters

{/* feishu-style:text-align:left */}
`temperature` represents the sampling temperature. Higher values (such as 0.8) will make the output more random, while lower values (such as 0.2) will make the output more deterministic.

{/* feishu-style:text-align:left */}
`top_p` represents the probability threshold for nucleus sampling, used to control the diversity of text generated by the model. The higher the value, the greater the diversity of the generated text.

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

In thinking mode, the `mimo-v2.6-flash`, `mimo-v2.6-pro`, `mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro` and `mimo-v2.5` models do not support customizing the `temperature` and `top_p` parameters. Even if these parameters are passed in, the actual effective values will be forcibly set by the model to its recommended default values of `1.0` and `0.95`.

</div>

{/* feishu-style:text-align:left */}
The default values and parameter ranges of `temperature` and `top_p` for different models are as follows:

<table>
<colgroup>
<col style="width: 40.47%" />
<col style="width: 28.45%" />
<col style="width: 31.08%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Model Name**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**temperature**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**top_p**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`</span></td>
<td><ul><li>Default value: 1.0</li><li>Range: [0, 1.5]</li></ul></td>
<td><ul><li>Default value: 0.95</li><li>Range: [0.01, 1.0]</li></ul></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`</span></td>
<td><ul><li>Default value: 1.0</li><li>Range: [0, 1.5]</li></ul></td>
<td><ul><li>Default value: 0.95</li><li>Range: [0.01, 1.0]</li></ul></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro-ultraspeed`</span></td>
<td><ul><li>Default value: 1.0</li><li>Range: [0, 1.5]</li></ul></td>
<td><ul><li>Default value: 0.95</li><li>Range: [0.01, 1.0]</li></ul></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-pro`(to be deprecated)</span></td>
<td><ul><li>Default value: 1.0</li><li>Range: [0, 1.5]</li></ul></td>
<td><ul><li>Default value: 0.95</li><li>Range: [0.01, 1.0]</li></ul></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5`(to be deprecated)</span></td>
<td><ul><li>Default value: 1.0</li><li>Range: [0, 1.5]</li></ul></td>
<td><ul><li>Default value: 0.95</li><li>Range: [0.01, 1.0]</li></ul></td>
</tr>
</tbody>
</table>


--- DOCUMENT: Error Codes ---
URL: https://mimo.mi.com/static/docs/api/guidance/error-codes.md

# Error Codes

{/* feishu-style:text-align:left */}
When using API calls to the MiMo model, common error codes and solutions are as follows:

<table>
<colgroup>
<col style="width: 20.6%" />
<col style="width: 35.88%" />
<col style="width: 43.519999999999996%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Error Code**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Causes**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Solutions**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">400 - Invalid Format</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Invalid request format</span></td>
<td><ul><li>Check if the JSON format is correct</li><li>Check if all required parameters are included</li><li>Check if parameter values are within the valid range</li><li>Check if the message format meets the interface requirements</li><li>Check if the model exists</li><li>Check if the fields are entered correctly</li><li>Check multimodal file input for compliance with format, size and other restrictions.</li><li>Check if multimodal file input is publicly accessible</li><li>In multi-turn conversations under thinking mode, the `reasoning_content` field must be fully passed back to the API.</li></ul></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">401 - Authentication Fails</span></td>
<td><ul><li>Missing or invalid API Key, or incorrect Authorization request header format</li><li>API Key that mixes Token Plan and Pay-as-you-go API</li></ul></td>
<td><ul><li>Check if the API key and request header format are correct</li><li>Check if a dedicated Base URL and API Key are used when using the Token Plan</li></ul></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">402 - Insufficient Balance</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Insufficient account balance</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Check your account balance and recharge in a timely manner</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">403 - Forbidden Access </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">The service is currently not available in the current region, or the API Key has been restricted by risk control </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Create a new API Key and pay attention to the security of input content</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">404 - Not Found</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">The requested endpoint or model does not support image input capability</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Verify that the model / endpoint being used supports image input capability</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">421 - Content Filter</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Content moderation and blocking</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Avoid entering unsafe or sensitive content</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">429 - Too Many Requests</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Requests are too frequent, or the quota of Token Plan has been exhausted</span></td>
<td><ul><li>Implement exponential backoff and retry logic, or reduce the request frequency</li><li>Upgrade the Token Plan package or switch to pay-as-you-go API</li></ul></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">500 - Server Error</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Our server encounters an issue</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Please try again later, or contact us for resolution</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">503 - Server Overloaded</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">The server is overloaded due to high traffic</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Please try again later</span></td>
</tr>
</tbody>
</table>


--- DOCUMENT: OpenAI Chat Completion API ---
URL: https://mimo.mi.com/static/docs/api/chat/openai-api.md

# OpenAI Chat Completions API Compatibility

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

Please refer to [Speech Synthesis (MiMo-TTS Series) - OpenAI API Compatibility](https://mimo.mi.com/docs/en-US/api/audio/tts) for the protocol documentation of the MiMo-TTS series models.

</div>

## Request Address

```bash
https://api.xiaomimimo.com/v1/chat/completions
```

## Request Headers

{/* feishu-style:text-align:left */}
The API supports the following two authentication methods. Please choose one and add it to the request headers:

<Tab>
  <TabItem label={`API Key Authentication`}>

```json
api-key: $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
  <TabItem label={`Bearer Authentication`}>

```json
Authorization: Bearer $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
</Tab>

## Request body 

<InlineSchemaV2 schema={`[
  {
    "name": "messages",
    "type": "array",
    "isBold": true,
    "required": true,
    "description": "The current conversation message list.",
    "children": [
      {
        "name": "Developer message",
        "type": "object",
        "isBold": false,
        "description": "Developer-provided instructions that the model should follow, regardless of messages sent by the user.",
        "children": [
          {
            "name": "content",
            "type": [
              "string",
              "array"
            ],
            "isBold": true,
            "required": true,
            "description": "The contents of the developer message.",
            "children": [
              {
                "name": "Text content",
                "type": "string",
                "isBold": false,
                "description": "The contents of the developer message."
              },
              {
                "name": "Array of content parts",
                "type": "array",
                "isBold": false,
                "description": "An array of content parts with a defined type. For developer messages, only type <code class=\\"schema-inline-code\\">text</code> is supported.",
                "children": [
                  {
                    "name": "Text content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The text content."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part."
                      }
                    ]
                  }
                ]
              }
            ]
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The role of the message author.<br />Available options: <code class=\\"schema-inline-code\\">developer</code>"
          },
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "required": false,
            "description": "An optional name for the participant. Provides the model information to differentiate between participants of the same role."
          }
        ]
      },
      {
        "name": "System message",
        "type": "object",
        "isBold": false,
        "description": "Developer-provided instructions that the model should follow, regardless of messages sent by the user.",
        "children": [
          {
            "name": "content",
            "type": [
              "string",
              "array"
            ],
            "isBold": true,
            "required": true,
            "description": "The contents of the system message.",
            "children": [
              {
                "name": "Text content",
                "type": "string",
                "isBold": false,
                "description": "The contents of the system message."
              },
              {
                "name": "Array of content parts",
                "type": "array",
                "isBold": false,
                "description": "An array of content parts with a defined type. For system messages, only type <code class=\\"schema-inline-code\\">text</code> is supported.",
                "children": [
                  {
                    "name": "Text content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The text content."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part."
                      }
                    ]
                  }
                ]
              }
            ]
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Role of the message author.<br />Available options: <code class=\\"schema-inline-code\\">system</code>"
          },
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "required": false,
            "description": "An optional name for the participant. Provides the model information to differentiate between participants of the same role."
          }
        ]
      },
      {
        "name": "User message",
        "type": "object",
        "isBold": false,
        "description": "Messages sent by an end user, containing prompts or additional context information.",
        "children": [
          {
            "name": "content",
            "type": [
              "string",
              "array"
            ],
            "isBold": true,
            "required": true,
            "description": "The contents of the user message.",
            "children": [
              {
                "name": "Text content",
                "type": "string",
                "isBold": false,
                "description": "The text contents of the message."
              },
              {
                "name": "Array of content parts",
                "type": "array",
                "isBold": false,
                "description": "An array of content parts with a defined type. Supported options differ based on the model being used to generate the response. Can contain text, image, audio or video inputs.<br /><blockquote class=\\"schema-blockquote\\">Currently, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models support image, audio or video input.</blockquote>",
                "children": [
                  {
                    "name": "Text content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The text content."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part."
                      }
                    ]
                  },
                  {
                    "name": "Image content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "image_url",
                        "type": "object",
                        "isBold": true,
                        "required": true,
                        "children": [
                          {
                            "name": "url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Either a URL of the image or the base64 encoded image data."
                          }
                        ]
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part.<br />Available options: <code class=\\"schema-inline-code\\">image_url</code>"
                      }
                    ]
                  },
                  {
                    "name": "Audio content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "input_audio",
                        "type": "object",
                        "isBold": true,
                        "required": true,
                        "children": [
                          {
                            "name": "data",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Either a URL of the audio or the base64 encoded audio data."
                          }
                        ]
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part.<br />Available options: <code class=\\"schema-inline-code\\">input_audio</code>"
                      }
                    ]
                  },
                  {
                    "name": "Video content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "video_url",
                        "type": "object",
                        "isBold": true,
                        "required": true,
                        "children": [
                          {
                            "name": "url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Either a URL of the video or the base64 encoded video data."
                          }
                        ]
                      },
                      {
                        "name": "fps",
                        "type": "number",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "2",
                        "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                      },
                      {
                        "name": "media_resolution",
                        "type": "string",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "default",
                        "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part.<br />Available options: <code class=\\"schema-inline-code\\">video_url</code>"
                      }
                    ]
                  }
                ]
              }
            ]
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Role of the message author.<br />Available options: <code class=\\"schema-inline-code\\">user</code>"
          },
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "required": false,
            "description": "An optional name for the participant. Provides model information to differentiate between participants of the same role."
          }
        ]
      },
      {
        "name": "Assistant message",
        "type": "object",
        "isBold": false,
        "description": "Messages sent by the model in response to user messages.",
        "children": [
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Role of the message author.<br />Available options: <code class=\\"schema-inline-code\\">assistant</code>"
          },
          {
            "name": "content",
            "type": [
              "string",
              "array"
            ],
            "isBold": true,
            "required": false,
            "description": "The contents of the assistant message. Required unless <code class=\\"schema-inline-code\\">tool_calls</code> is specified.",
            "children": [
              {
                "name": "Text content",
                "type": "string",
                "isBold": false,
                "description": "The contents of the assistant message."
              },
              {
                "name": "Array of content parts",
                "type": "array",
                "isBold": false,
                "description": "An array of content parts with a defined type. Can be one or more of type <code class=\\"schema-inline-code\\">text</code>.",
                "children": [
                  {
                    "name": "Text content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The text content."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part."
                      }
                    ]
                  }
                ]
              }
            ]
          },
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "required": false,
            "description": "An optional name for the participant. Provides model information to differentiate between participants of the same role."
          },
          {
            "name": "tool_calls",
            "type": "array",
            "isBold": true,
            "required": false,
            "description": "The tool calls generated by the model, such as function calls.",
            "children": [
              {
                "name": "Function tool call",
                "type": "object",
                "isBold": false,
                "description": "A call to a function tool created by the model.",
                "children": [
                  {
                    "name": "function",
                    "type": "object",
                    "isBold": true,
                    "required": true,
                    "description": "The function that the model called.",
                    "children": [
                      {
                        "name": "arguments",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The arguments to call the function with, as generated by the model in JSON format. Note that the model does not always generate valid JSON, and may hallucinate parameters not defined by your function schema. Validate the arguments in your code before calling your function."
                      },
                      {
                        "name": "name",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The name of the function to call."
                      }
                    ]
                  },
                  {
                    "name": "id",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The ID of the tool call."
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "Tool type. Currently, only <code class=\\"schema-inline-code\\">function</code> is supported."
                  }
                ]
              }
            ]
          }
        ]
      },
      {
        "name": "Tool message",
        "type": "object",
        "isBold": false,
        "children": [
          {
            "name": "content",
            "type": [
              "string",
              "array"
            ],
            "isBold": true,
            "required": true,
            "description": "The contents of the tool message.",
            "children": [
              {
                "name": "Text content",
                "type": "string",
                "isBold": false,
                "description": "The contents of the tool message."
              },
              {
                "name": "Array of content parts",
                "type": "array",
                "isBold": false,
                "description": "An array of content parts with a defined type. For tool messages, text, image, audio and video are supported.<br /><blockquote class=\\"schema-blockquote\\">Currently, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models support image, audio or video input.</blockquote>",
                "children": [
                  {
                    "name": "Text content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The text content."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part."
                      }
                    ]
                  },
                  {
                    "name": "Image content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "image_url",
                        "type": "object",
                        "isBold": true,
                        "required": true,
                        "children": [
                          {
                            "name": "url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Either a URL of the image or the base64 encoded image data."
                          }
                        ]
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part.<br />Available options: <code class=\\"schema-inline-code\\">image_url</code>"
                      }
                    ]
                  },
                  {
                    "name": "Audio content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "input_audio",
                        "type": "object",
                        "isBold": true,
                        "required": true,
                        "children": [
                          {
                            "name": "data",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Either a URL of the audio or the base64 encoded audio data."
                          }
                        ]
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part.<br />Available options: <code class=\\"schema-inline-code\\">input_audio</code>"
                      }
                    ]
                  },
                  {
                    "name": "Video content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "video_url",
                        "type": "object",
                        "isBold": true,
                        "required": true,
                        "children": [
                          {
                            "name": "url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Either a URL of the video or the base64 encoded video data."
                          }
                        ]
                      },
                      {
                        "name": "fps",
                        "type": "number",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "2",
                        "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                      },
                      {
                        "name": "media_resolution",
                        "type": "string",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "default",
                        "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part.<br />Available options: <code class=\\"schema-inline-code\\">video_url</code>"
                      }
                    ]
                  }
                ]
              }
            ]
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Role of the message author.<br />Available options: <code class=\\"schema-inline-code\\">tool</code>"
          },
          {
            "name": "tool_call_id",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Tool call that this message is responding to."
          }
        ]
      }
    ]
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "required": true,
    "description": "Model ID is used to generate the response.<br />Available options: <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code>"
  },
  {
    "name": "frequency_penalty",
    "type": [
      "number",
      "null"
    ],
    "isBold": true,
    "required": false,
    "defaultValue": "0",
    "description": "Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.<br />Required range: <code class=\\"schema-inline-code\\">[-2.0, 2.0]</code>"
  },
  {
    "name": "max_completion_tokens",
    "type": [
      "integer",
      "null"
    ],
    "isBold": true,
    "required": false,
    "description": "An upper bound for the number of tokens that can be generated for a completion, including visible output tokens and reasoning tokens.<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5</code>: default <code class=\\"schema-inline-code\\">32768</code></li></ul>Required range: <code class=\\"schema-inline-code\\">[1, 131072]</code>"
  },
  {
    "name": "presence_penalty",
    "type": [
      "number",
      "null"
    ],
    "isBold": true,
    "required": false,
    "defaultValue": "0",
    "description": "Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.<br />Required range: <code class=\\"schema-inline-code\\">[-2.0, 2.0]</code>"
  },
  {
    "name": "response_format",
    "type": "object",
    "isBold": true,
    "required": false,
    "description": "An object specifying the format that the model must output.",
    "children": [
      {
        "name": "Text",
        "type": "object",
        "isBold": false,
        "description": "Default response format. Used to generate text responses.",
        "children": [
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The type of response format being defined. Always <code class=\\"schema-inline-code\\">text</code>."
          }
        ]
      },
      {
        "name": "JSON object",
        "type": "object",
        "isBold": false,
        "description": "JSON object response format. Note that the model will not generate JSON without a system or user message instructing it to do so.",
        "children": [
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The type of response format being defined. Always <code class=\\"schema-inline-code\\">json_object</code>."
          }
        ]
      }
    ]
  },
  {
    "name": "stop",
    "type": [
      "string",
      "array",
      "null"
    ],
    "isBold": true,
    "required": false,
    "defaultValue": "null",
    "description": "Up to 4 sequences where the API will stop generating further tokens. The returned text will not contain the stop sequence."
  },
  {
    "name": "stream",
    "type": [
      "boolean",
      "null"
    ],
    "isBold": true,
    "required": false,
    "defaultValue": "false",
    "description": "If set to true, the model response data will be streamed to the client as it is generated using server-sent events."
  },
  {
    "name": "thinking",
    "type": "object",
    "isBold": true,
    "required": false,
    "description": "This parameter is used to control whether the model enables the chain of thought.<br /><blockquote class=\\"schema-blockquote\\">Note: During the multi-turn tool calls process in thinking mode, the model returns a <code class=\\"schema-inline-code\\">reasoning_content</code> field alongside <code class=\\"schema-inline-code\\">tool_calls</code>. To continue the conversation, it is recommended to keep all previous <code class=\\"schema-inline-code\\">reasoning_content</code> in the <code class=\\"schema-inline-code\\">messages</code> array for each subsequent request to achieve the best performance.</blockquote><blockquote class=\\"schema-blockquote\\">In thinking mode, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models do not support customizing the <code class=\\"schema-inline-code\\">temperature</code> and <code class=\\"schema-inline-code\\">top_p</code> parameters. Even if these parameters are passed in, the actual effective values will be forcibly set by the model to its recommended default values of <code class=\\"schema-inline-code\\">1.0</code> and <code class=\\"schema-inline-code\\">0.95</code>.</blockquote>",
    "children": [
      {
        "name": "type",
        "type": "string",
        "isBold": true,
        "required": true,
        "description": "Whether to enable the chain of thought.<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code>: default <code class=\\"schema-inline-code\\">enabled</code></li></ul>Available options: <code class=\\"schema-inline-code\\">enabled</code>, <code class=\\"schema-inline-code\\">disabled</code>"
      }
    ]
  },
  {
    "name": "temperature",
    "type": "number",
    "isBold": true,
    "required": false,
    "description": "What sampling temperature to use, between 0 and 1.5. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic. We generally recommend altering this or <code class=\\"schema-inline-code\\">top_p</code> but not both.<br /><blockquote class=\\"schema-blockquote\\">In thinking mode, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models do not support customizing the <code class=\\"schema-inline-code\\">temperature</code> parameter. Even if this parameter is passed in, it will be forcibly overridden and take effect with the model's recommended default value of <code class=\\"schema-inline-code\\">1.0</code>.</blockquote><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code>: default <code class=\\"schema-inline-code\\">1.0</code></li></ul>Required range: <code class=\\"schema-inline-code\\">[0, 1.5]</code>"
  },
  {
    "name": "tool_choice",
    "type": "string",
    "isBold": true,
    "required": false,
    "description": "Controls how the model selects a tool.<br /><blockquote class=\\"schema-blockquote\\">Note: When a value other than <code class=\\"schema-inline-code\\">auto</code> is passed to <code class=\\"schema-inline-code\\">tool_choice</code>, the backend will remove this field by default, and the model response behavior will still be equivalent to the <code class=\\"schema-inline-code\\">auto</code> mode (this logic is subject to future adjustments).</blockquote>Available options: <code class=\\"schema-inline-code\\">auto</code>"
  },
  {
    "name": "tools",
    "type": "array",
    "isBold": true,
    "required": false,
    "description": "A list of tools the model may call. You can provide function tools.<br /><blockquote class=\\"schema-blockquote\\">Note: During the multi-turn tool calls process in thinking mode, the model returns a <code class=\\"schema-inline-code\\">reasoning_content</code> field alongside <code class=\\"schema-inline-code\\">tool_calls</code>. To continue the conversation, it is recommended to keep all previous <code class=\\"schema-inline-code\\">reasoning_content</code> in the <code class=\\"schema-inline-code\\">messages</code> array for each subsequent request to achieve the best performance.</blockquote>",
    "children": [
      {
        "name": "Function tool",
        "type": "object",
        "isBold": false,
        "description": "A function tool that can be used to generate a response.",
        "children": [
          {
            "name": "function",
            "type": "object",
            "isBold": true,
            "required": true,
            "children": [
              {
                "name": "name",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The name of the tool function. Must be <code class=\\"schema-inline-code\\">a-z</code>, <code class=\\"schema-inline-code\\">A-Z</code>, <code class=\\"schema-inline-code\\">0-9</code>, or contain underscores (<code class=\\"schema-inline-code\\">_</code>) and dashes (<code class=\\"schema-inline-code\\">-</code>), with a maximum length of 64.<br />Required string length: <code class=\\"schema-inline-code\\">1 - 64</code>"
              },
              {
                "name": "description",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "A description of what the function does, used by the model to choose when and how to call the function."
              },
              {
                "name": "parameters",
                "type": "object",
                "isBold": true,
                "required": false,
                "description": "The parameters the functions accept, described as a JSON Schema object.<br />Omitting <code class=\\"schema-inline-code\\">parameters</code> defines a function with an empty parameter list."
              },
              {
                "name": "strict",
                "type": "boolean",
                "isBold": true,
                "required": false,
                "defaultValue": "false",
                "description": "Whether to enable strict schema adherence when generating the function call. If set to true, the model will follow the exact schema defined in the <code class=\\"schema-inline-code\\">parameters</code> field. Only a subset of JSON Schema is supported when <code class=\\"schema-inline-code\\">strict</code> is <code class=\\"schema-inline-code\\">true</code>."
              }
            ]
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Tool type. Currently, only <code class=\\"schema-inline-code\\">function</code> is supported."
          }
        ]
      },
      {
        "name": "Web search tool",
        "type": "object",
        "isBold": false,
        "description": "A web search tool that can be used to generate a response.For details, please refer to <a target=\\"_blank\\" rel=\\"noopener noreferrer\\" href=\\"https://platform.xiaomimimo.com/#/docs/usage-guide/tool-calling/web-search\\">Web Search</a>.<br /><blockquote class=\\"schema-blockquote\\">Note：<a target=\\"_blank\\" rel=\\"noopener noreferrer\\" href=\\"https://platform.xiaomimimo.com/#/console/plugin\\">Web Search plugin</a> must be activated before use.</blockquote>",
        "children": [
          {
            "name": "user_location",
            "type": "object",
            "isBold": true,
            "required": false,
            "children": [
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "approximate"
              },
              {
                "name": "country",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "country"
              },
              {
                "name": "region",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "region"
              },
              {
                "name": "city",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "city"
              },
              {
                "name": "district",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "district"
              },
              {
                "name": "longitude",
                "type": "long",
                "isBold": true,
                "required": false,
                "description": "longitude<strong> </strong>"
              },
              {
                "name": "latitude",
                "type": "long",
                "isBold": true,
                "required": false,
                "description": "latitude"
              }
            ]
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Tool type. Currently, only <code class=\\"schema-inline-code\\">web_search</code> is supported."
          },
          {
            "name": "force_search",
            "type": "boolean",
            "isBold": true,
            "required": false,
            "defaultValue": "false",
            "description": "Whether to enable forced search. <code class=\\"schema-inline-code\\">true</code> for forced search, <code class=\\"schema-inline-code\\">false</code> for the model to decide whether search is needed."
          },
          {
            "name": "max_keyword",
            "type": "integer",
            "isBold": true,
            "required": false,
            "defaultValue": "5",
            "description": "Limit the maximum number of keywords that can be used in a single search.<br />Required range: <code class=\\"schema-inline-code\\">[1, 50]</code>"
          },
          {
            "name": "limit",
            "type": "integer",
            "isBold": true,
            "required": false,
            "defaultValue": "5",
            "description": "Limit the maximum number of results returned by a single search operation.<br />Required range: <code class=\\"schema-inline-code\\">[1, 50]</code>"
          }
        ]
      }
    ]
  },
  {
    "name": "top_p",
    "type": "number",
    "isBold": true,
    "required": false,
    "defaultValue": "0.95",
    "description": "The probability threshold for nucleus sampling, which controls the diversity of the text that the model generates. A higher <code class=\\"schema-inline-code\\">top_p</code> value results in more diverse text. A lower <code class=\\"schema-inline-code\\">top_p</code> value results in more deterministic text.<br />Because both <code class=\\"schema-inline-code\\">temperature</code> and <code class=\\"schema-inline-code\\">top_p</code> control the diversity of the generated text, we recommend that you set only one of them.<br /><blockquote class=\\"schema-blockquote\\">In thinking mode, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code> models do not support customizing the <code class=\\"schema-inline-code\\">top_p</code> parameter. Even if this parameter is passed in, it will be forcibly overridden and take effect with the model's recommended default value of <code class=\\"schema-inline-code\\">0.95</code>.</blockquote>Required range: <code class=\\"schema-inline-code\\">[0.01, 1.0]</code>"
  }
]`} />

## Chat response object (non-streaming output) 

<InlineSchemaV2 schema={`[
  {
    "name": "choices",
    "type": "array",
    "isBold": true,
    "description": "A list of chat completion choices.",
    "children": [
      {
        "name": "finish_reason",
        "type": "string",
        "isBold": true,
        "description": "The reason the model stopped generating tokens. This will be <code class=\\"schema-inline-code\\">stop</code> if the model hit a natural stop point or a provided stop sequence, <code class=\\"schema-inline-code\\">length</code> if the maximum number of tokens specified in the request was reached, <code class=\\"schema-inline-code\\">tool_calls</code> if the model called a tool, <code class=\\"schema-inline-code\\">content_filter</code> if content was omitted due to a flag from our content filters, <code class=\\"schema-inline-code\\">repetition_truncation</code> if the model detects repetition."
      },
      {
        "name": "index",
        "type": "integer",
        "isBold": true,
        "description": "The index of the choice in the list of choices."
      },
      {
        "name": "message",
        "type": "object",
        "isBold": true,
        "description": "A chat completion message generated by the model.",
        "children": [
          {
            "name": "content",
            "type": "string",
            "isBold": true,
            "description": "The contents of the message."
          },
          {
            "name": "reasoning_content",
            "type": "string",
            "isBold": true,
            "description": "The reasoning contents of the assistant message, before the final answer."
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "description": "The role of the author of this message."
          },
          {
            "name": "tool_calls",
            "type": "array",
            "isBold": true,
            "description": "After a function call is initiated, the model returns the tool to be called and the parameters that are Required for the call. This parameter can contain one or more tool response objects.",
            "children": [
              {
                "name": "Function tool call",
                "type": "object",
                "isBold": false,
                "description": "A call to a function tool created by the model.",
                "children": [
                  {
                    "name": "function",
                    "type": "object",
                    "isBold": true,
                    "description": "The function that the model called.",
                    "children": [
                      {
                        "name": "arguments",
                        "type": "string",
                        "isBold": true,
                        "description": "The arguments to call the function with, as generated by the model in JSON format. Note      that the model does not always generate valid JSON, and may hallucinate parameters not    defined by your function schema. Validate the arguments in your code before calling your function."
                      },
                      {
                        "name": "name",
                        "type": "string",
                        "isBold": true,
                        "description": "The name of the function to call."
                      }
                    ]
                  },
                  {
                    "name": "id",
                    "type": "string",
                    "isBold": true,
                    "description": "The ID of the tool call."
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "description": "The type of the tool. Currently, only <code class=\\"schema-inline-code\\">function</code> is supported."
                  }
                ]
              }
            ]
          },
          {
            "name": "annotations",
            "type": "array",
            "isBold": true,
            "description": "After web search, the model returns annotations for all referenced URLs.",
            "children": [
              {
                "name": "web_search tool call",
                "type": "object",
                "isBold": false,
                "description": "A call to a web search tool created by the model.",
                "children": [
                  {
                    "name": "logo_url",
                    "type": "string",
                    "isBold": true,
                    "description": "Logo url."
                  },
                  {
                    "name": "publish_time",
                    "type": "string",
                    "isBold": true,
                    "description": "Publish time."
                  },
                  {
                    "name": "site_name",
                    "type": "string",
                    "isBold": true,
                    "description": "Site name."
                  },
                  {
                    "name": "summary",
                    "type": "string",
                    "isBold": true,
                    "description": "Summary."
                  },
                  {
                    "name": "title",
                    "type": "string",
                    "isBold": true,
                    "description": "Title."
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "description": "Type."
                  },
                  {
                    "name": "url",
                    "type": "string",
                    "isBold": true,
                    "description": "Url."
                  }
                ]
              }
            ]
          },
          {
            "name": "error_message",
            "type": "string",
            "isBold": true,
            "description": "Error message of web search."
          }
        ]
      }
    ]
  },
  {
    "name": "created",
    "type": "integer",
    "isBold": true,
    "description": "The Unix timestamp (in seconds) of when the chat completion was created."
  },
  {
    "name": "id",
    "type": "string",
    "isBold": true,
    "description": "A unique identifier for the chat completion."
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "description": "The model to generate the completion."
  },
  {
    "name": "object",
    "type": "string",
    "isBold": true,
    "description": "The object type, which is always <code class=\\"schema-inline-code\\">chat.completion</code>."
  },
  {
    "name": "usage",
    "type": [
      "object",
      "null"
    ],
    "isBold": true,
    "description": "Usage statistics for the completion request.",
    "children": [
      {
        "name": "completion_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the generated completion."
      },
      {
        "name": "prompt_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the prompt."
      },
      {
        "name": "total_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Total number of tokens used in the request (prompt + completion)."
      },
      {
        "name": "completion_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in a completion.",
        "children": [
          {
            "name": "reasoning_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Tokens generated by the model for reasoning."
          }
        ]
      },
      {
        "name": "prompt_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in the prompt.",
        "children": [
          {
            "name": "cached_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Number of tokens served from cache."
          },
          {
            "name": "audio_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Audio input tokens present in the prompt."
          },
          {
            "name": "image_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Image input tokens present in the prompt."
          },
          {
            "name": "video_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Video input tokens present in the prompt."
          }
        ]
      },
      {
        "name": "web_search_usage",
        "type": "object",
        "isBold": true,
        "description": "Detailed usage of the web search API.",
        "children": [
          {
            "name": "tool_usage",
            "type": "integer",
            "isBold": true,
            "description": "Number of API calls in web search."
          },
          {
            "name": "page_usage",
            "type": "integer",
            "isBold": true,
            "description": "Number of web pages returned by the web search API."
          }
        ]
      }
    ]
  }
]`} />

## Chat response chunk object (streaming output) 
<InlineSchemaV2 schema={`[
  {
    "name": "choices",
    "type": "array",
    "isBold": true,
    "description": "A list of chat completion choices.",
    "children": [
      {
        "name": "delta",
        "type": "object",
        "isBold": true,
        "description": "A chat completion delta generated by streamed model responses.",
        "children": [
          {
            "name": "content",
            "type": "string",
            "isBold": true,
            "description": "The contents of the chunk message."
          },
          {
            "name": "reasoning_content",
            "type": "string",
            "isBold": true,
            "description": "The reasoning contents of the assistant message, before the final answer."
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "description": "The role of the author of this message."
          },
          {
            "name": "tool_calls",
            "type": "array",
            "isBold": true,
            "description": "The tools to be called by the model and the parameters Required for the calls. It can contain one or more tool response objects.",
            "children": [
              {
                "name": "index",
                "type": "integer",
                "isBold": true,
                "description": "The index of the called tool in the <code class=\\"schema-inline-code\\">tool_calls</code> list, starting from 0."
              },
              {
                "name": "function",
                "type": "object",
                "isBold": true,
                "description": "The function to be called.",
                "children": [
                  {
                    "name": "arguments",
                    "type": "string",
                    "isBold": true,
                    "description": "The arguments to call the function with, as generated by the model in JSON format. Note that the model does not always generate valid JSON, and may hallucinate parameters not defined by your function schema. Validate the arguments in your code before calling your function."
                  },
                  {
                    "name": "name",
                    "type": "string",
                    "isBold": true,
                    "description": "The name of the function to call."
                  }
                ]
              },
              {
                "name": "id",
                "type": "string",
                "isBold": true,
                "description": "The ID of the tool call."
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "description": "The type of the tool. Currently, only <code class=\\"schema-inline-code\\">function</code> is supported."
              }
            ]
          },
          {
            "name": "annotations",
            "type": "array",
            "isBold": true,
            "description": "After web search, the model returns annotations for all referenced URLs.",
            "children": [
              {
                "name": "web_search tool call",
                "type": "object",
                "isBold": false,
                "description": "A call to a web search tool created by the model.",
                "children": [
                  {
                    "name": "logo_url",
                    "type": "string",
                    "isBold": true,
                    "description": "Logo url."
                  },
                  {
                    "name": "publish_time",
                    "type": "string",
                    "isBold": true,
                    "description": "Publish time."
                  },
                  {
                    "name": "site_name",
                    "type": "string",
                    "isBold": true,
                    "description": "Site name."
                  },
                  {
                    "name": "summary",
                    "type": "string",
                    "isBold": true,
                    "description": "Summary."
                  },
                  {
                    "name": "title",
                    "type": "string",
                    "isBold": true,
                    "description": "Title."
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "description": "Type."
                  },
                  {
                    "name": "url",
                    "type": "string",
                    "isBold": true,
                    "description": "Url."
                  }
                ]
              }
            ]
          },
          {
            "name": "error_message",
            "type": "string",
            "isBold": true,
            "description": "Error message of web search."
          }
        ]
      },
      {
        "name": "finish_reason",
        "type": [
          "string",
          "null"
        ],
        "isBold": true,
        "description": "The reason the model stopped generating tokens. This will be <code class=\\"schema-inline-code\\">stop</code> if the model hit a natural stop point or a provided stop sequence, <code class=\\"schema-inline-code\\">length</code> if the maximum number of tokens specified in the request was reached, <code class=\\"schema-inline-code\\">tool_calls</code> if the model called a tool, <code class=\\"schema-inline-code\\">content_filter</code> if content was omitted due to a flag from our content filters, <code class=\\"schema-inline-code\\">repetition_truncation</code> if the model detects repetition."
      },
      {
        "name": "index",
        "type": "integer",
        "isBold": true,
        "description": "The index of the choice in the list of choices."
      }
    ]
  },
  {
    "name": "created",
    "type": "integer",
    "isBold": true,
    "description": "The Unix timestamp (in seconds) of when the chat completion was created. Each chunk has the same timestamp."
  },
  {
    "name": "id",
    "type": "string",
    "isBold": true,
    "description": "A unique identifier for the chat completion.  Each chunk has the same ID."
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "description": "The model to generate the completion."
  },
  {
    "name": "object",
    "type": "string",
    "isBold": true,
    "description": "The object type, which is always <code class=\\"schema-inline-code\\">chat.completion.chunk</code>."
  },
  {
    "name": "usage",
    "type": [
      "object",
      "null"
    ],
    "isBold": true,
    "description": "Usage statistics for the completion request.",
    "children": [
      {
        "name": "completion_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the generated completion."
      },
      {
        "name": "prompt_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the prompt."
      },
      {
        "name": "total_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Total number of tokens used in the request (prompt + completion)."
      },
      {
        "name": "completion_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in a completion.",
        "children": [
          {
            "name": "reasoning_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Tokens generated by the model for reasoning."
          }
        ]
      },
      {
        "name": "prompt_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in the prompt.",
        "children": [
          {
            "name": "cached_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Number of tokens served from cache."
          },
          {
            "name": "audio_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Audio input tokens present in the prompt."
          },
          {
            "name": "image_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Image input tokens present in the prompt."
          },
          {
            "name": "video_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Video input tokens present in the prompt."
          }
        ]
      },
      {
        "name": "web_search_usage",
        "type": "object",
        "isBold": true,
        "description": "Detailed usage of the web search API.",
        "children": [
          {
            "name": "tool_usage",
            "type": "integer",
            "isBold": true,
            "description": "Number of API calls in web search."
          },
          {
            "name": "page_usage",
            "type": "integer",
            "isBold": true,
            "description": "Number of web pages returned by the web search API."
          }
        ]
      }
    ]
  }
]`} />


--- DOCUMENT: OpenAI Responses API ---
URL: https://mimo.mi.com/static/docs/api/chat/responses.md

# OpenAI Responses API Compatibility

{/* feishu-style:text-align:left */}
MiMo provides a calling interface compatible with the OpenAI Responses API format. This document covers request parameters, response schemas and code examples.

{/* feishu-style:text-align:left */}
**Compatibility Notes & Limitations:** 

{/* feishu-style:text-align:left */}
This interface aligns with the OpenAI Responses API specification to facilitate quick integration for developers. Only the parameters documented here will be processed normally; undefined parameters will be filtered out and may cause request errors. See below for specific behavioral differences.

- **Incompatible parameters**: Fields such as `background`, `previous_response_id`, and `context_management` are not currently supported. Carrying these parameters in a request will be ignored or trigger an error.

- **Reasoning level control**: `reasoning.effort` controls model reasoning. `none` disables thinking; Every other level enables thinking with identical behavior — The reasoning intensity is not differentiated at this stage. 

## Request Address

```bash
https://api.xiaomimimo.com/v1/responses
```

## Request Headers

{/* feishu-style:text-align:left */}
The API supports the following two authentication methods. Please choose one and add it to the request headers:

<Tab>
  <TabItem label={`API Key Authentication`}>

```json
api-key: $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
  <TabItem label={`Bearer Authentication`}>

```json
Authorization: Bearer $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
</Tab>

## Request Body

<InlineSchemaV2 schema={`[
  {
    "name": "input",
    "type": [
      "string",
      "array"
    ],
    "isBold": true,
    "required": true,
    "description": "Text, image, audio, video inputs to the model, used to generate a response.",
    "children": [
      {
        "name": "TextInput",
        "type": "string",
        "isBold": false,
        "description": "A text input to the model, equivalent to a text input with the <code class=\\"schema-inline-code\\">user</code> role."
      },
      {
        "name": "InputItemList",
        "type": "array",
        "isBold": false,
        "description": "A list of one or many input items to the model, containing different content types.",
        "children": [
          {
            "name": "EasyInputMessage",
            "type": "object",
            "isBold": false,
            "description": "A message input to the model with a role indicating instruction following hierarchy. Instructions given with the <code class=\\"schema-inline-code\\">developer</code> or <code class=\\"schema-inline-code\\">system</code> role take precedence over instructions given with the <code class=\\"schema-inline-code\\">user</code> role.",
            "children": [
              {
                "name": "content",
                "type": [
                  "string",
                  "array"
                ],
                "isBold": true,
                "required": true,
                "description": "Text, image, audio and video inputs to the model, used to generate a response.",
                "children": [
                  {
                    "name": "TextInput",
                    "type": "string",
                    "isBold": false,
                    "description": "A text input to the model."
                  },
                  {
                    "name": "ResponseInputMessageContentList",
                    "type": "array",
                    "isBold": false,
                    "description": "A list of one or many input items to the model, containing different content types.<br /><blockquote class=\\"schema-blockquote\\">Currently, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models support image, audio or video input.</blockquote>",
                    "children": [
                      {
                        "name": "ResponseInputText",
                        "type": "object",
                        "isBold": false,
                        "description": "A text input to the model.",
                        "children": [
                          {
                            "name": "text",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The text input to the model."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_text</code>"
                          }
                        ]
                      },
                      {
                        "name": "ResponseInputImage",
                        "type": "object",
                        "isBold": false,
                        "description": "An image input to the model.",
                        "children": [
                          {
                            "name": "image_url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_image</code>"
                          }
                        ]
                      },
                      {
                        "name": "ResponseInputAudio",
                        "type": "object",
                        "isBold": false,
                        "description": "An audio input to the model.",
                        "children": [
                          {
                            "name": "audio_url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The URL of the audio to be sent to the model. A fully qualified URL or base64 encoded audio in a data URL."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_audio</code>"
                          }
                        ]
                      },
                      {
                        "name": "ResponseInputVideo",
                        "type": "object",
                        "isBold": false,
                        "description": "A video input to the model.",
                        "children": [
                          {
                            "name": "video_url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The URL of the video to be sent to the model. A fully qualified URL or base64 encoded video in a data URL."
                          },
                          {
                            "name": "fps",
                            "type": "number",
                            "isBold": true,
                            "required": false,
                            "defaultValue": "2",
                            "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                          },
                          {
                            "name": "media_resolution",
                            "type": "string",
                            "isBold": true,
                            "required": false,
                            "defaultValue": "default",
                            "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_video</code>"
                          }
                        ]
                      }
                    ]
                  }
                ]
              },
              {
                "name": "role",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The role of the message input.<br />Available options: <code class=\\"schema-inline-code\\">user</code>, <code class=\\"schema-inline-code\\">assistant</code>, <code class=\\"schema-inline-code\\">system</code>, <code class=\\"schema-inline-code\\">developer</code>"
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The type of the message input.<br />Available options: <code class=\\"schema-inline-code\\">message</code>"
              }
            ]
          },
          {
            "name": "Message",
            "type": "object",
            "isBold": false,
            "description": "A message input to the model with a role indicating instruction following hierarchy.",
            "children": [
              {
                "name": "content",
                "type": "array",
                "isBold": true,
                "required": true,
                "description": "A list of one or many input items to the model, containing different content types.<br /><blockquote class=\\"schema-blockquote\\">Currently, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models support image, audio or video input.</blockquote>",
                "children": [
                  {
                    "name": "ResponseInputText",
                    "type": "object",
                    "isBold": false,
                    "description": "A text input to the model.",
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The text input to the model."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_text</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputImage",
                    "type": "object",
                    "isBold": false,
                    "description": "An image input to the model.",
                    "children": [
                      {
                        "name": "image_url",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_image</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputAudio",
                    "type": "object",
                    "isBold": false,
                    "description": "An audio input to the model.",
                    "children": [
                      {
                        "name": "audio_url",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The URL of the audio to be sent to the model. A fully qualified URL or base64 encoded audio in a data URL."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_audio</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputVideo",
                    "type": "object",
                    "isBold": false,
                    "description": "A video input to the model.",
                    "children": [
                      {
                        "name": "video_url",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The URL of the video to be sent to the model. A fully qualified URL or base64 encoded video in a data URL."
                      },
                      {
                        "name": "fps",
                        "type": "number",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "2",
                        "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                      },
                      {
                        "name": "media_resolution",
                        "type": "string",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "default",
                        "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_video</code>"
                      }
                    ]
                  }
                ]
              },
              {
                "name": "role",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The role of the message input.<br />Available options: <code class=\\"schema-inline-code\\">user</code>, <code class=\\"schema-inline-code\\">system</code>, <code class=\\"schema-inline-code\\">developer</code>"
              },
              {
                "name": "status",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The status of item. Populated when items are returned via API.<br />Available options: <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">completed</code>"
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The type of the message input.<br />Available options: <code class=\\"schema-inline-code\\">message</code>"
              }
            ]
          },
          {
            "name": "ResponseOutputMessage",
            "type": "object",
            "isBold": false,
            "description": "An output message from the model.",
            "children": [
              {
                "name": "id",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The unique ID of the output message."
              },
              {
                "name": "content",
                "type": "array",
                "isBold": true,
                "required": true,
                "description": "The content of the output message.",
                "children": [
                  {
                    "name": "ResponseOutputText",
                    "type": "object",
                    "isBold": false,
                    "description": "A text output from the model.",
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The text output from the model."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the output text.<br />Available options: <code class=\\"schema-inline-code\\">output_text</code>"
                      }
                    ]
                  }
                ]
              },
              {
                "name": "role",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The role of the output message.<br />Available options: <code class=\\"schema-inline-code\\">assistant</code>"
              },
              {
                "name": "status",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The status of the message input. Populated when input items are returned via API.<br />Available options: <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">completed</code>"
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The type of the output message.<br />Available options: <code class=\\"schema-inline-code\\">message</code>"
              }
            ]
          },
          {
            "name": "FunctionCall",
            "type": "object",
            "isBold": false,
            "description": "A tool call to run a function.",
            "children": [
              {
                "name": "arguments",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "A JSON string of the arguments to pass to the function."
              },
              {
                "name": "call_id",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The unique ID of the function tool call generated by the model."
              },
              {
                "name": "name",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The name of the function to run."
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The type of the function tool call.<br />Available options: <code class=\\"schema-inline-code\\">function_call</code>"
              },
              {
                "name": "id",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The unique ID of the function tool call."
              },
              {
                "name": "namespace",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The namespace of the function to run."
              },
              {
                "name": "status",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The status of the item. Populated when items are returned via API.<br />Available options: <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">completed</code>"
              }
            ]
          },
          {
            "name": "FunctionCallOutput",
            "type": "object",
            "isBold": false,
            "description": "The output of a function tool call.",
            "children": [
              {
                "name": "call_id",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The unique ID of the function tool call generated by the model."
              },
              {
                "name": "output",
                "type": [
                  "string",
                  "array"
                ],
                "isBold": true,
                "required": true,
                "description": "Text, image, audio or video output of the function tool call.",
                "children": [
                  {
                    "name": "StringOutput",
                    "type": "string",
                    "isBold": false,
                    "description": "A JSON string of the output of the function tool call."
                  },
                  {
                    "name": "OutputContentList",
                    "type": "array",
                    "isBold": false,
                    "description": "An array of content outputs for the function tool call.<br /><blockquote class=\\"schema-blockquote\\">Currently, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models support image, audio or video input.</blockquote>",
                    "children": [
                      {
                        "name": "ResponseInputTextContent",
                        "type": "object",
                        "isBold": false,
                        "description": "A text input to the model.",
                        "children": [
                          {
                            "name": "text",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The text input to the model."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_text</code>"
                          }
                        ]
                      },
                      {
                        "name": "ResponseInputImageContent",
                        "type": "object",
                        "isBold": false,
                        "description": "An image input to the model.",
                        "children": [
                          {
                            "name": "image_url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_image</code>"
                          }
                        ]
                      },
                      {
                        "name": "ResponseInputAudioContent",
                        "type": "object",
                        "isBold": false,
                        "description": "An audio input to the model.",
                        "children": [
                          {
                            "name": "audio_url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The URL of the audio to be sent to the model. A fully qualified URL or base64 encoded audio in a data URL."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_audio</code>"
                          }
                        ]
                      },
                      {
                        "name": "ResponseInputVideoContent",
                        "type": "object",
                        "isBold": false,
                        "description": "A video input to the model.",
                        "children": [
                          {
                            "name": "video_url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The URL of the video to be sent to the model. A fully qualified URL or base64 encoded video in a data URL."
                          },
                          {
                            "name": "fps",
                            "type": "number",
                            "isBold": true,
                            "required": false,
                            "defaultValue": "2",
                            "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                          },
                          {
                            "name": "media_resolution",
                            "type": "string",
                            "isBold": true,
                            "required": false,
                            "defaultValue": "default",
                            "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_video</code>"
                          }
                        ]
                      }
                    ]
                  }
                ]
              },
              {
                "name": "id",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The unique ID of the function tool call output. Populated when this item is returned via API."
              },
              {
                "name": "status",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The status of the item. Populated when items are returned via API.<br />Available options: <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">completed</code>"
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The type of the function tool call output.<br />Available options: <code class=\\"schema-inline-code\\">function_call_output</code>"
              }
            ]
          },
          {
            "name": "AgentMessage[Multi-Agent]",
            "type": "object",
            "isBold": false,
            "description": "A message routed between agents.",
            "children": [
              {
                "name": "author",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The sending agent identity."
              },
              {
                "name": "content",
                "type": "array",
                "isBold": true,
                "required": true,
                "description": "Plaintext, image, audio, video or encrypted content sent between agents.<br /><blockquote class=\\"schema-blockquote\\">Currently, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models support image, audio or video input.</blockquote>",
                "children": [
                  {
                    "name": "ResponseInputTextContent",
                    "type": "object",
                    "isBold": false,
                    "description": "A text input to the model.",
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The text input to the model."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_text</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputImageContent",
                    "type": "object",
                    "isBold": false,
                    "description": "An image input to the model.",
                    "children": [
                      {
                        "name": "image_url",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_image</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputAudioContent",
                    "type": "object",
                    "isBold": false,
                    "description": "An audio input to the model.",
                    "children": [
                      {
                        "name": "audio_url",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The URL of the audio to be sent to the model. A fully qualified URL or base64 encoded audio in a data URL."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_audio</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputVideoContent",
                    "type": "object",
                    "isBold": false,
                    "description": "A video input to the model.",
                    "children": [
                      {
                        "name": "video_url",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The URL of the video to be sent to the model. A fully qualified URL or base64 encoded video in a data URL."
                      },
                      {
                        "name": "fps",
                        "type": "number",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "2",
                        "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                      },
                      {
                        "name": "media_resolution",
                        "type": "string",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "default",
                        "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_video</code>"
                      }
                    ]
                  },
                  {
                    "name": "EncryptedContent",
                    "type": "object",
                    "isBold": false,
                    "description": "Opaque encrypted content that Responses API decrypts inside trusted model execution. For MiMo, the content is plaintext and no decryption is performed.",
                    "children": [
                      {
                        "name": "encrypted_content",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "Opaque encrypted content."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">encrypted_content</code>"
                      }
                    ]
                  }
                ]
              },
              {
                "name": "recipient",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The destination agent identity."
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The item type.<br />Available options: <code class=\\"schema-inline-code\\">agent_message</code>"
              },
              {
                "name": "id",
                "type": [
                  "string",
                  "null"
                ],
                "isBold": true,
                "required": false,
                "description": "The unique ID of this agent message item."
              },
              {
                "name": "agent",
                "type": [
                  "object",
                  "null"
                ],
                "isBold": true,
                "required": false,
                "description": "The agent that produced this item.",
                "children": [
                  {
                    "name": "agent_name",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The canonical name of the agent that produced this item."
                  }
                ]
              }
            ]
          },
          {
            "name": "AdditionalTools",
            "type": "object",
            "isBold": false,
            "children": [
              {
                "name": "role",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The role that provided the additional tools.<br />Available options: <code class=\\"schema-inline-code\\">developer</code>"
              },
              {
                "name": "tools",
                "type": "array",
                "isBold": true,
                "required": true,
                "description": "A list of additional tools made available at this item.",
                "children": [
                  {
                    "name": "Function",
                    "type": "object",
                    "isBold": false,
                    "description": "Defines a function in your own code the model can choose to call.",
                    "children": [
                      {
                        "name": "name",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The name of the tool function. Must be <code class=\\"schema-inline-code\\">a-z</code>, <code class=\\"schema-inline-code\\">A-Z</code>, <code class=\\"schema-inline-code\\">0-9</code>, or contain underscores (<code class=\\"schema-inline-code\\">_</code>) and dashes (<code class=\\"schema-inline-code\\">-</code>), with a maximum length of 64.<br />Required string length: <code class=\\"schema-inline-code\\">1 - 64</code>"
                      },
                      {
                        "name": "parameters",
                        "type": "object",
                        "isBold": true,
                        "required": true,
                        "description": "A JSON schema object describing the parameters of the function."
                      },
                      {
                        "name": "strict",
                        "type": "boolean",
                        "isBold": true,
                        "required": true,
                        "defaultValue": "false",
                        "description": "Whether to enable strict schema adherence when generating the function call."
                      },
                      {
                        "name": "description",
                        "type": [
                          "string",
                          "null"
                        ],
                        "isBold": true,
                        "required": false,
                        "description": "A description of the function. Used by the model to determine whether or not to call the function."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the tool.<br />Available options: <code class=\\"schema-inline-code\\">function</code>"
                      }
                    ]
                  },
                  {
                    "name": "Custom",
                    "type": "object",
                    "isBold": false,
                    "description": "A custom tool that processes input using a specified format.",
                    "children": [
                      {
                        "name": "name",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The name of the custom tool, used to identify it in tool calls."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the custom tool.<br />Available options: <code class=\\"schema-inline-code\\">custom</code>"
                      },
                      {
                        "name": "description",
                        "type": "string",
                        "isBold": true,
                        "required": false,
                        "description": "Optional description of the custom tool, used to provide more context."
                      },
                      {
                        "name": "format",
                        "type": "object",
                        "isBold": true,
                        "required": false,
                        "description": "The input format for the custom tool. Default is unconstrained text.",
                        "children": [
                          {
                            "name": "CustomToolInputFormat",
                            "type": "object",
                            "isBold": false,
                            "children": [
                              {
                                "name": "Text",
                                "type": "object",
                                "isBold": false,
                                "description": "Unconstrained free-form text.",
                                "children": [
                                  {
                                    "name": "type",
                                    "type": "string",
                                    "isBold": true,
                                    "required": true,
                                    "description": "Unconstrained text format.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
                                  }
                                ]
                              },
                              {
                                "name": "Grammar",
                                "type": "object",
                                "isBold": false,
                                "description": "A grammar defined by the user.",
                                "children": [
                                  {
                                    "name": "definition",
                                    "type": "string",
                                    "isBold": true,
                                    "required": true,
                                    "description": "The grammar definition."
                                  },
                                  {
                                    "name": "syntax",
                                    "type": "string",
                                    "isBold": true,
                                    "required": true,
                                    "description": "The syntax of the grammar definition.<br />Available options: <code class=\\"schema-inline-code\\">lark</code>, <code class=\\"schema-inline-code\\">regex</code>"
                                  },
                                  {
                                    "name": "type",
                                    "type": "string",
                                    "isBold": true,
                                    "required": true,
                                    "description": "Grammar format.<br />Available options: <code class=\\"schema-inline-code\\">grammar</code>"
                                  }
                                ]
                              }
                            ]
                          }
                        ]
                      }
                    ]
                  },
                  {
                    "name": "Namespace",
                    "type": "object",
                    "isBold": false,
                    "description": "Groups function tools under a shared namespace.",
                    "children": [
                      {
                        "name": "description",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "A description of the namespace shown to the model."
                      },
                      {
                        "name": "name",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The namespace name used in tool calls.<br />MinLength: <code class=\\"schema-inline-code\\">1</code>"
                      },
                      {
                        "name": "tools",
                        "type": "array",
                        "isBold": true,
                        "required": true,
                        "description": "The function tools available inside this namespace.",
                        "children": [
                          {
                            "name": "Function",
                            "type": "object",
                            "isBold": false,
                            "children": [
                              {
                                "name": "name",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "Required string length: <code class=\\"schema-inline-code\\">1 - 64</code>"
                              },
                              {
                                "name": "type",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "Available options: <code class=\\"schema-inline-code\\">function</code>"
                              },
                              {
                                "name": "description",
                                "type": [
                                  "string",
                                  "null"
                                ],
                                "isBold": true,
                                "required": false
                              },
                              {
                                "name": "parameters",
                                "type": "object",
                                "isBold": true,
                                "required": false
                              },
                              {
                                "name": "strict",
                                "type": "boolean",
                                "isBold": true,
                                "required": false
                              }
                            ]
                          },
                          {
                            "name": "Custom",
                            "type": "object",
                            "isBold": false,
                            "description": "A custom tool that processes input using a specified format.",
                            "children": [
                              {
                                "name": "name",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "The name of the custom tool, used to identify it in tool calls."
                              },
                              {
                                "name": "type",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "The type of the custom tool.<br />Available options: <code class=\\"schema-inline-code\\">custom</code>"
                              },
                              {
                                "name": "description",
                                "type": "string",
                                "isBold": true,
                                "required": false,
                                "description": "Optional description of the custom tool, used to provide more context."
                              },
                              {
                                "name": "format",
                                "type": "object",
                                "isBold": true,
                                "required": false,
                                "description": "The input format for the custom tool. Default is unconstrained text.",
                                "children": [
                                  {
                                    "name": "CustomToolInputFormat",
                                    "type": "object",
                                    "isBold": false,
                                    "children": [
                                      {
                                        "name": "Text",
                                        "type": "object",
                                        "isBold": false,
                                        "description": "Unconstrained free-form text.",
                                        "children": [
                                          {
                                            "name": "type",
                                            "type": "string",
                                            "isBold": true,
                                            "required": true,
                                            "description": "Unconstrained text format.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
                                          }
                                        ]
                                      },
                                      {
                                        "name": "Grammar",
                                        "type": "object",
                                        "isBold": false,
                                        "description": "A grammar defined by the user.",
                                        "children": [
                                          {
                                            "name": "definition",
                                            "type": "string",
                                            "isBold": true,
                                            "required": true,
                                            "description": "The grammar definition."
                                          },
                                          {
                                            "name": "syntax",
                                            "type": "string",
                                            "isBold": true,
                                            "required": true,
                                            "description": "The syntax of the grammar definition.<br />Available options: <code class=\\"schema-inline-code\\">lark</code>, <code class=\\"schema-inline-code\\">regex</code>"
                                          },
                                          {
                                            "name": "type",
                                            "type": "string",
                                            "isBold": true,
                                            "required": true,
                                            "description": "Grammar format.<br />Available options: <code class=\\"schema-inline-code\\">grammar</code>"
                                          }
                                        ]
                                      }
                                    ]
                                  }
                                ]
                              }
                            ]
                          }
                        ]
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the tool.<br />Available options: <code class=\\"schema-inline-code\\">namespace</code>"
                      }
                    ]
                  }
                ]
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The item type.<br />Available options: <code class=\\"schema-inline-code\\">additional_tools</code>"
              },
              {
                "name": "id",
                "type": [
                  "string",
                  "null"
                ],
                "isBold": true,
                "required": false,
                "description": "The unique ID of this additional tools item."
              }
            ]
          },
          {
            "name": "Reasoning",
            "type": "object",
            "isBold": false,
            "description": "A description of the chain of thought used by a reasoning model while generating a response.",
            "children": [
              {
                "name": "id",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The unique identifier of the reasoning content."
              },
              {
                "name": "content",
                "type": "array",
                "isBold": true,
                "required": false,
                "description": "Reasoning text content.",
                "children": [
                  {
                    "name": "text",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The reasoning text from the model."
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The type of the object.<br />Available options: <code class=\\"schema-inline-code\\">reasoning_text</code>"
                  }
                ]
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The type of the object.<br />Available options: <code class=\\"schema-inline-code\\">reasoning</code>"
              },
              {
                "name": "status",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The status of the item. Populated when items are returned via API.<br />Available options: <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">completed</code>"
              }
            ]
          },
          {
            "name": "CustomToolCallOutput",
            "type": "object",
            "isBold": false,
            "description": "The output of a custom tool call from your code, being sent back to the model.",
            "children": [
              {
                "name": "call_id",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The call ID, used to map this custom tool call output to a custom tool call."
              },
              {
                "name": "output",
                "type": [
                  "string",
                  "array"
                ],
                "isBold": true,
                "required": true,
                "description": "The output from the custom tool call generated by your code. Can be a string or a list of output content.",
                "children": [
                  {
                    "name": "StringOutput",
                    "type": "string",
                    "isBold": false,
                    "description": "A string of the output of the custom tool call."
                  },
                  {
                    "name": "OutputContentList",
                    "type": "array",
                    "isBold": false,
                    "description": "An array of content outputs for the function tool call.<br /><blockquote class=\\"schema-blockquote\\">Currently, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models support image, audio or video input.</blockquote>",
                    "children": [
                      {
                        "name": "ResponseInputText",
                        "type": "object",
                        "isBold": false,
                        "description": "A text input to the model.",
                        "children": [
                          {
                            "name": "text",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The text input to the model."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_text</code>"
                          }
                        ]
                      },
                      {
                        "name": "ResponseInputImage",
                        "type": "object",
                        "isBold": false,
                        "description": "An image input to the model.",
                        "children": [
                          {
                            "name": "image_url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_image</code>"
                          }
                        ]
                      },
                      {
                        "name": "ResponseInputAudio",
                        "type": "object",
                        "isBold": false,
                        "description": "An audio input to the model.",
                        "children": [
                          {
                            "name": "audio_url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The URL of the audio to be sent to the model. A fully qualified URL or base64 encoded audio in a data URL."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_audio</code>"
                          }
                        ]
                      },
                      {
                        "name": "ResponseInputVideo",
                        "type": "object",
                        "isBold": false,
                        "description": "A video input to the model.",
                        "children": [
                          {
                            "name": "video_url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The URL of the video to be sent to the model. A fully qualified URL or base64 encoded video in a data URL."
                          },
                          {
                            "name": "fps",
                            "type": "number",
                            "isBold": true,
                            "required": false,
                            "defaultValue": "2",
                            "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                          },
                          {
                            "name": "media_resolution",
                            "type": "string",
                            "isBold": true,
                            "required": false,
                            "defaultValue": "default",
                            "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_video</code>"
                          }
                        ]
                      }
                    ]
                  }
                ]
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "description": "The type of the custom tool call output.<br />Available options: <code class=\\"schema-inline-code\\">custom_tool_call_output</code>"
              },
              {
                "name": "id",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The unique ID of the custom tool call output."
              }
            ]
          },
          {
            "name": "CustomToolCall",
            "type": "object",
            "isBold": false,
            "description": "A call to a custom tool created by the model.",
            "children": [
              {
                "name": "call_id",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "An identifier used to map this custom tool call to a tool call output."
              },
              {
                "name": "input",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The input for the custom tool call generated by the model."
              },
              {
                "name": "name",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The name of the custom tool being called."
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The type of the custom tool call.<br />Available options: <code class=\\"schema-inline-code\\">custom_tool_call</code>"
              },
              {
                "name": "id",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The unique ID of the custom tool call."
              },
              {
                "name": "namespace",
                "type": "string",
                "isBold": true,
                "required": false,
                "description": "The namespace of the custom tool being called."
              }
            ]
          }
        ]
      }
    ]
  },
  {
    "name": "instructions",
    "type": "string",
    "isBold": true,
    "required": false,
    "description": "A system (or developer) message inserted into the model's context."
  },
  {
    "name": "max_output_tokens",
    "type": "integer",
    "isBold": true,
    "required": false,
    "description": "An upper bound for the number of tokens that can be generated for a response, including visible output tokens and reasoning tokens.<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5</code>: default <code class=\\"schema-inline-code\\">32768</code></li></ul>Required range: <code class=\\"schema-inline-code\\">[1, 131072]</code>"
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "required": true,
    "description": "Model ID used to generate the response.<br />Available options: <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code>"
  },
  {
    "name": "stream",
    "type": "boolean",
    "isBold": true,
    "required": false,
    "defaultValue": "false",
    "description": "If set to true, the model response data will be streamed to the client as it is generated using server-sent events."
  },
  {
    "name": "reasoning",
    "type": "object",
    "isBold": true,
    "required": false,
    "description": "Configuration options for reasoning models.<br /><blockquote class=\\"schema-blockquote\\">Note: During the multi-turn tool calls process in thinking mode, the model returns the reasoning content alongside the tool calls field. To continue the conversation, it is recommended to keep all previous reasoning content in the <code class=\\"schema-inline-code\\">input</code> array for each subsequent request to achieve the best performance.</blockquote><blockquote class=\\"schema-blockquote\\">In thinking mode, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models do not support customizing the <code class=\\"schema-inline-code\\">temperature</code> and <code class=\\"schema-inline-code\\">top_p</code> parameters. Even if these parameters are passed in, the actual effective values will be forcibly set by the model to its recommended default values of <code class=\\"schema-inline-code\\">1.0</code> and <code class=\\"schema-inline-code\\">0.95</code>.</blockquote>",
    "children": [
      {
        "name": "effort",
        "type": "string",
        "isBold": true,
        "required": true,
        "description": "Constrains effort on reasoning for reasoning models. Reducing reasoning effort can result in faster responses and fewer tokens used on reasoning in a response.<br /><blockquote class=\\"schema-blockquote\\">Custom tuning of reasoning effort is currently unsupported. When set to <code class=\\"schema-inline-code\\">none</code>, reasoning is disabled; all other valid values map to enabled reasoning. The value <code class=\\"schema-inline-code\\">minimal</code> is mapped to <code class=\\"schema-inline-code\\">low</code>. Values <code class=\\"schema-inline-code\\">xhigh</code>, <code class=\\"schema-inline-code\\">max</code> and <code class=\\"schema-inline-code\\">ultra</code> are mapped to <code class=\\"schema-inline-code\\">high</code>.</blockquote><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code>: default <code class=\\"schema-inline-code\\">enabled</code></li></ul>Available options: <code class=\\"schema-inline-code\\">none</code>, <code class=\\"schema-inline-code\\">minimal</code>, <code class=\\"schema-inline-code\\">low</code>, <code class=\\"schema-inline-code\\">medium</code>, <code class=\\"schema-inline-code\\">high</code>, <code class=\\"schema-inline-code\\">xhigh</code>, <code class=\\"schema-inline-code\\">max</code>, <code class=\\"schema-inline-code\\">ultra</code>"
      }
    ]
  },
  {
    "name": "temperature",
    "type": "number",
    "isBold": true,
    "required": false,
    "description": "What sampling temperature to use, between 0 and 1.5. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic. We generally recommend altering this or <code class=\\"schema-inline-code\\">top_p</code> but not both.<br /><blockquote class=\\"schema-blockquote\\">In thinking mode, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models do not support customizing the <code class=\\"schema-inline-code\\">temperature</code> parameter. Even if this parameter is passed in, it will be forcibly overridden and take effect with the model's recommended default value of <code class=\\"schema-inline-code\\">1.0</code>.</blockquote><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code>: default <code class=\\"schema-inline-code\\">1.0</code></li></ul>Required range: <code class=\\"schema-inline-code\\">[0, 1.5]</code>"
  },
  {
    "name": "text",
    "type": "object",
    "isBold": true,
    "required": false,
    "description": "Configuration options for a text response from the model. Can be plain text or structured JSON data.",
    "children": [
      {
        "name": "format",
        "type": "object",
        "isBold": true,
        "required": false,
        "description": "An object specifying the format that the model must output. The default format is <code class=\\"schema-inline-code\\">{ &quot;type&quot;: &quot;text&quot; }</code> with no additional options.",
        "children": [
          {
            "name": "ResponseFormatText",
            "type": "object",
            "isBold": false,
            "description": "Default response format. Used to generate text responses.",
            "children": [
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The type of response format being defined.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
              }
            ]
          },
          {
            "name": "ResponseFormatJSONObject",
            "type": "object",
            "isBold": false,
            "description": "JSON object response format.<br /><blockquote class=\\"schema-blockquote\\">Note: If the output of JSON is not required by system instructions or user instructions, the model will not actively generate JSON.</blockquote>",
            "children": [
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "required": true,
                "description": "The type of response format being defined.<br />Available options: <code class=\\"schema-inline-code\\">json_object</code>"
              }
            ]
          }
        ]
      }
    ]
  },
  {
    "name": "tool_choice",
    "type": "string",
    "isBold": true,
    "required": false,
    "description": "Controls how the model calls tools.<br /><blockquote class=\\"schema-blockquote\\">Note: When a value other than <code class=\\"schema-inline-code\\">auto</code> is passed to <code class=\\"schema-inline-code\\">tool_choice</code>, the backend will remove this field by default, and the model response behavior will still be equivalent to the <code class=\\"schema-inline-code\\">auto</code> mode (this logic is subject to future adjustments).</blockquote>Available options: <code class=\\"schema-inline-code\\">auto</code>"
  },
  {
    "name": "tools",
    "type": "array",
    "isBold": true,
    "required": false,
    "description": "An array of tools the model may call while generating a response. You can specify which tool to use by setting the <code class=\\"schema-inline-code\\">tool_choice</code> parameter.<br /><blockquote class=\\"schema-blockquote\\">Note: During the multi-turn tool calls process in thinking mode, the model returns the reasoning content alongside the tool calls field. To continue the conversation, it is recommended to keep all previous reasoning content in the <code class=\\"schema-inline-code\\">input</code> array for each subsequent request to achieve the best performance.</blockquote>",
    "children": [
      {
        "name": "Function",
        "type": "object",
        "isBold": false,
        "description": "Defines a function in your own code the model can choose to call.",
        "children": [
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The name of the tool function. Must be <code class=\\"schema-inline-code\\">a-z</code>, <code class=\\"schema-inline-code\\">A-Z</code>, <code class=\\"schema-inline-code\\">0-9</code>, or contain underscores (<code class=\\"schema-inline-code\\">_</code>) and dashes (<code class=\\"schema-inline-code\\">-</code>), with a maximum length of 64.<br />Required string length: <code class=\\"schema-inline-code\\">1 - 64</code>"
          },
          {
            "name": "parameters",
            "type": "object",
            "isBold": true,
            "required": true,
            "description": "A JSON schema object describing the parameters of the function."
          },
          {
            "name": "strict",
            "type": "boolean",
            "isBold": true,
            "required": true,
            "defaultValue": "false",
            "description": "Whether to enable strict schema adherence when generating the function call."
          },
          {
            "name": "description",
            "type": "string",
            "isBold": true,
            "required": false,
            "description": "A description of the function. Used by the model to determine whether or not to call the function."
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The type of the tool.<br />Available options: <code class=\\"schema-inline-code\\">function</code>"
          }
        ]
      },
      {
        "name": "Custom",
        "type": "object",
        "isBold": false,
        "description": "A custom tool that processes input using a specified format.",
        "children": [
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The name of the custom tool, used to identify it in tool calls."
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The type of the custom tool.<br />Available options: <code class=\\"schema-inline-code\\">custom</code>"
          },
          {
            "name": "description",
            "type": "string",
            "isBold": true,
            "required": false,
            "description": "Optional description of the custom tool, used to provide more context."
          },
          {
            "name": "format",
            "type": "object",
            "isBold": true,
            "required": false,
            "description": "The input format for the custom tool. Default is unconstrained text.",
            "children": [
              {
                "name": "CustomToolInputFormat",
                "type": "object",
                "isBold": false,
                "children": [
                  {
                    "name": "Text",
                    "type": "object",
                    "isBold": false,
                    "description": "Unconstrained free-form text.",
                    "children": [
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "Unconstrained text format.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
                      }
                    ]
                  },
                  {
                    "name": "Grammar",
                    "type": "object",
                    "isBold": false,
                    "description": "A grammar defined by the user.",
                    "children": [
                      {
                        "name": "definition",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The grammar definition."
                      },
                      {
                        "name": "syntax",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The syntax of the grammar definition.<br />Available options: <code class=\\"schema-inline-code\\">lark</code>, <code class=\\"schema-inline-code\\">regex</code>"
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "Grammar format.<br />Available options: <code class=\\"schema-inline-code\\">grammar</code>"
                      }
                    ]
                  }
                ]
              }
            ]
          }
        ]
      },
      {
        "name": "Namespace",
        "type": "object",
        "isBold": false,
        "description": "Groups function tools under a shared namespace.",
        "children": [
          {
            "name": "description",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "A description of the namespace shown to the model."
          },
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The namespace name used in tool calls.<br />MinLength: <code class=\\"schema-inline-code\\">1</code>"
          },
          {
            "name": "tools",
            "type": "array",
            "isBold": true,
            "required": true,
            "description": "The function tools available inside this namespace.",
            "children": [
              {
                "name": "Function",
                "type": "object",
                "isBold": false,
                "children": [
                  {
                    "name": "name",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "Required string length: <code class=\\"schema-inline-code\\">1 - 64</code>"
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "Available options: <code class=\\"schema-inline-code\\">function</code>"
                  },
                  {
                    "name": "description",
                    "type": [
                      "string",
                      "null"
                    ],
                    "isBold": true,
                    "required": false
                  },
                  {
                    "name": "parameters",
                    "type": "object",
                    "isBold": true,
                    "required": false
                  },
                  {
                    "name": "strict",
                    "type": "boolean",
                    "isBold": true,
                    "required": false
                  }
                ]
              },
              {
                "name": "Custom",
                "type": "object",
                "isBold": false,
                "description": "A custom tool that processes input using a specified format.",
                "children": [
                  {
                    "name": "name",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The name of the custom tool, used to identify it in tool calls."
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The type of the custom tool.<br />Available options: <code class=\\"schema-inline-code\\">custom</code>"
                  },
                  {
                    "name": "description",
                    "type": "string",
                    "isBold": true,
                    "required": false,
                    "description": "Optional description of the custom tool, used to provide more context."
                  },
                  {
                    "name": "format",
                    "type": "object",
                    "isBold": true,
                    "required": false,
                    "description": "The input format for the custom tool. Default is unconstrained text.",
                    "children": [
                      {
                        "name": "CustomToolInputFormat",
                        "type": "object",
                        "isBold": false,
                        "children": [
                          {
                            "name": "Text",
                            "type": "object",
                            "isBold": false,
                            "description": "Unconstrained free-form text.",
                            "children": [
                              {
                                "name": "type",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "Unconstrained text format.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
                              }
                            ]
                          },
                          {
                            "name": "Grammar",
                            "type": "object",
                            "isBold": false,
                            "description": "A grammar defined by the user.",
                            "children": [
                              {
                                "name": "definition",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "The grammar definition."
                              },
                              {
                                "name": "syntax",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "The syntax of the grammar definition.<br />Available options: <code class=\\"schema-inline-code\\">lark</code>, <code class=\\"schema-inline-code\\">regex</code>"
                              },
                              {
                                "name": "type",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "Grammar format.<br />Available options: <code class=\\"schema-inline-code\\">grammar</code>"
                              }
                            ]
                          }
                        ]
                      }
                    ]
                  }
                ]
              }
            ]
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The type of the tool.<br />Available options: <code class=\\"schema-inline-code\\">namespace</code>"
          }
        ]
      }
    ]
  },
  {
    "name": "top_p",
    "type": "number",
    "isBold": true,
    "required": false,
    "defaultValue": "0.95",
    "description": "An alternative to sampling with temperature, called nucleus sampling. We generally recommend altering this or <code class=\\"schema-inline-code\\">temperature</code> but not both.<br /><blockquote class=\\"schema-blockquote\\">In thinking mode, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models do not support customizing the <code class=\\"schema-inline-code\\">top_p</code> parameter. Even if this parameter is passed in, it will be forcibly overridden and take effect with the model's recommended default value of <code class=\\"schema-inline-code\\">0.95</code>.</blockquote>Required range: <code class=\\"schema-inline-code\\">[0.01, 1.0]</code>"
  }
]`} />

## Response Object (non-streaming output) 

<InlineSchemaV2 schema={`[
  {
    "name": "id",
    "type": "string",
    "isBold": true,
    "description": "Unique identifier for this Response."
  },
  {
    "name": "created_at",
    "type": "number",
    "isBold": true,
    "description": "Unix timestamp (in seconds) of when this Response was created."
  },
  {
    "name": "error",
    "type": "object",
    "isBold": true,
    "description": "An error object returned when the model fails to generate a Response.",
    "children": [
      {
        "name": "ResponseError",
        "type": "object",
        "isBold": false,
        "children": [
          {
            "name": "code",
            "type": "string",
            "isBold": true,
            "description": "The error code for the response."
          },
          {
            "name": "message",
            "type": "string",
            "isBold": true,
            "description": "A human-readable description of the error."
          }
        ]
      }
    ]
  },
  {
    "name": "incomplete_details",
    "type": "object",
    "isBold": true,
    "description": "Details about why the response is incomplete.",
    "children": [
      {
        "name": "reason",
        "type": "string",
        "isBold": true,
        "description": "The reason why the response is incomplete.<br />Available options: <code class=\\"schema-inline-code\\">max_output_tokens</code>, <code class=\\"schema-inline-code\\">content_filter</code>"
      }
    ]
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "description": "Model ID used to generate the response."
  },
  {
    "name": "object",
    "type": "string",
    "isBold": true,
    "description": "Available options: <code class=\\"schema-inline-code\\">response</code>"
  },
  {
    "name": "output",
    "type": "array",
    "isBold": true,
    "description": "An array of content items generated by the model.<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\">The length and order of items in the <code class=\\"schema-inline-code\\">output</code> array is dependent on the model’s response.</li><li class=\\"schema-list-item\\">Rather than accessing the first item in the <code class=\\"schema-inline-code\\">output</code> array and assuming it’s an <code class=\\"schema-inline-code\\">assistant</code> message with the content generated by the model, you might consider using the <code class=\\"schema-inline-code\\">output_text</code> property where supported in SDKs.</li></ul>",
    "children": [
      {
        "name": "ResponseOutputMessage",
        "type": "object",
        "isBold": false,
        "description": "A message output from the model.",
        "children": [
          {
            "name": "id",
            "type": "string",
            "isBold": true,
            "description": "The unique ID of the output message."
          },
          {
            "name": "content",
            "type": "array",
            "isBold": true,
            "description": "The content of the output message.",
            "children": [
              {
                "name": "ResponseOutputText",
                "type": "object",
                "isBold": false,
                "description": "A text output from the model.",
                "children": [
                  {
                    "name": "text",
                    "type": "string",
                    "isBold": true,
                    "description": "The text output from the model."
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "description": "Available options: <code class=\\"schema-inline-code\\">output_text</code>"
                  }
                ]
              }
            ]
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "description": "The role of the output message.<br />Available options: <code class=\\"schema-inline-code\\">assistant</code>"
          },
          {
            "name": "status",
            "type": "string",
            "isBold": true,
            "description": "The status of the message.<br />Available options: <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">completed</code>"
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "Available options: <code class=\\"schema-inline-code\\">message</code>"
          }
        ]
      },
      {
        "name": "FunctionCall",
        "type": "object",
        "isBold": false,
        "description": "A tool call to a function tool.",
        "children": [
          {
            "name": "arguments",
            "type": "string",
            "isBold": true,
            "description": "The arguments that the model generated for the tool call, as a JSON string."
          },
          {
            "name": "call_id",
            "type": "string",
            "isBold": true,
            "description": "An identifier used when responding to the tool call with output."
          },
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "description": "The name of the tool to call."
          },
          {
            "name": "id",
            "type": "string",
            "isBold": true,
            "description": "The unique ID of the tool call."
          },
          {
            "name": "namespace",
            "type": "string",
            "isBold": true,
            "description": "The namespace of the function to run."
          },
          {
            "name": "status",
            "type": "string",
            "isBold": true,
            "description": "The status of the item.<br />Available options: <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">completed</code>"
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "Available options: <code class=\\"schema-inline-code\\">function_call</code>"
          }
        ]
      },
      {
        "name": "FunctionCallOutput",
        "type": "object",
        "isBold": false,
        "description": "The output of a function tool call.",
        "children": [
          {
            "name": "call_id",
            "type": "string",
            "isBold": true,
            "description": "The unique ID of the function tool call generated by the model."
          },
          {
            "name": "output",
            "type": [
              "string",
              "array"
            ],
            "isBold": true,
            "description": "The output from the function call generated by your code. Can be a string or a list of output content.",
            "children": [
              {
                "name": "StringOutput",
                "type": "string",
                "isBold": false,
                "description": "A string of the output of the function tool call."
              },
              {
                "name": "OutputContentList",
                "type": "array",
                "isBold": false,
                "description": "Text, image, audio or video output of the function call.",
                "children": [
                  {
                    "name": "ResponseInputText",
                    "type": "object",
                    "isBold": false,
                    "description": "A text input to the model.",
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The text input to the model."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_text</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputImage",
                    "type": "object",
                    "isBold": false,
                    "description": "An image input to the model.",
                    "children": [
                      {
                        "name": "image_url",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_image</code>."
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputAudioContent",
                    "type": "object",
                    "isBold": false,
                    "description": "An audio input to the model.",
                    "children": [
                      {
                        "name": "audio_url",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The URL of the audio to be sent to the model. A fully qualified URL or base64 encoded audio in a data URL."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_audio</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputVideoContent",
                    "type": "object",
                    "isBold": false,
                    "description": "A video input to the model.",
                    "children": [
                      {
                        "name": "video_url",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The URL of the video to be sent to the model. A fully qualified URL or base64 encoded video in a data URL."
                      },
                      {
                        "name": "fps",
                        "type": "number",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "2",
                        "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                      },
                      {
                        "name": "media_resolution",
                        "type": "string",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "default",
                        "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_video</code>"
                      }
                    ]
                  }
                ]
              }
            ]
          },
          {
            "name": "id",
            "type": "string",
            "isBold": true,
            "description": "The unique ID of the function tool call output. Populated when this item is returned via API."
          },
          {
            "name": "status",
            "type": "string",
            "isBold": true,
            "description": "The status of the item.<br />Available options: <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">completed</code>"
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the function tool call output.<br />Always <code class=\\"schema-inline-code\\">function_call_output</code>."
          }
        ]
      },
      {
        "name": "Reasoning",
        "type": "object",
        "isBold": false,
        "description": "A description of the chain of thought used by a reasoning model while generating a response. Be sure to include these items in your input to the Responses API for subsequent turns of a conversation if you are manually managing context.",
        "children": [
          {
            "name": "id",
            "type": "string",
            "isBold": true,
            "description": "The unique identifier of the reasoning content."
          },
          {
            "name": "content",
            "type": "array",
            "isBold": true,
            "description": "Reasoning text content.",
            "children": [
              {
                "name": "text",
                "type": "string",
                "isBold": true,
                "description": "The reasoning text from the model."
              },
              {
                "name": "type",
                "type": "string",
                "isBold": true,
                "description": "The type of the reasoning text.<br />Available options: <code class=\\"schema-inline-code\\">reasoning_text</code>"
              }
            ]
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the object.<br />Available options: <code class=\\"schema-inline-code\\">reasoning</code>"
          },
          {
            "name": "status",
            "type": "string",
            "isBold": true,
            "description": "The status of the item.<br />Available options: <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">completed</code>"
          }
        ]
      },
      {
        "name": "CustomToolCall",
        "type": "object",
        "isBold": false,
        "description": "A call to a custom tool created by the model.",
        "children": [
          {
            "name": "call_id",
            "type": "string",
            "isBold": true,
            "description": "An identifier used to map this custom tool call to a tool call output."
          },
          {
            "name": "input",
            "type": "string",
            "isBold": true,
            "description": "The input for the custom tool call generated by the model."
          },
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "description": "The name of the custom tool being called."
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the custom tool call.<br />Available options: <code class=\\"schema-inline-code\\">custom_tool_call</code>"
          },
          {
            "name": "id",
            "type": "string",
            "isBold": true,
            "description": "The unique ID of the custom tool call."
          },
          {
            "name": "namespace",
            "type": "string",
            "isBold": true,
            "description": "The namespace of the custom tool being called."
          }
        ]
      },
      {
        "name": "CustomToolCallOutput",
        "type": "object",
        "isBold": false,
        "description": "The output of a custom tool call from your code, being sent back to the model.",
        "children": [
          {
            "name": "id",
            "type": "string",
            "isBold": true,
            "description": "The unique ID of the custom tool call output."
          },
          {
            "name": "call_id",
            "type": "string",
            "isBold": true,
            "description": "The call ID, used to map this custom tool call output to a custom tool call."
          },
          {
            "name": "output",
            "type": [
              "string",
              "array"
            ],
            "isBold": true,
            "description": "The output from the custom tool call generated by your code. Can be a string or a list of output content.",
            "children": [
              {
                "name": "StringOutput",
                "type": "string",
                "isBold": false,
                "description": "A string of the output of the custom tool call."
              },
              {
                "name": "OutputContentList",
                "type": "array",
                "isBold": false,
                "description": "An array of content outputs for the function tool call.",
                "children": [
                  {
                    "name": "ResponseInputText",
                    "type": "object",
                    "isBold": false,
                    "description": "A text input to the model.",
                    "children": [
                      {
                        "name": "text",
                        "type": "string",
                        "isBold": true,
                        "description": "The text input to the model."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "description": "The type of the input item.<br />Available options: <code class=\\"schema-inline-code\\">input_text</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputImage",
                    "type": "object",
                    "isBold": false,
                    "description": "An image input to the model.",
                    "children": [
                      {
                        "name": "image_url",
                        "type": "string",
                        "isBold": true,
                        "description": "The URL of the image to be sent to the model. A fully qualified URL or base64 encoded image in a data URL."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_image</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputAudio",
                    "type": "object",
                    "isBold": false,
                    "description": "An audio input to the model.",
                    "children": [
                      {
                        "name": "audio_url",
                        "type": "string",
                        "isBold": true,
                        "description": "The URL of the audio to be sent to the model. A fully qualified URL or base64 encoded audio in a data URL."
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_audio</code>"
                      }
                    ]
                  },
                  {
                    "name": "ResponseInputVideo",
                    "type": "object",
                    "isBold": false,
                    "description": "A video input to the model.",
                    "children": [
                      {
                        "name": "video_url",
                        "type": "string",
                        "isBold": true,
                        "description": "The URL of the video to be sent to the model. A fully qualified URL or base64 encoded video in a data URL."
                      },
                      {
                        "name": "fps",
                        "type": "number",
                        "isBold": true,
                        "defaultValue": "2",
                        "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                      },
                      {
                        "name": "media_resolution",
                        "type": "string",
                        "isBold": true,
                        "defaultValue": "default",
                        "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "description": "The type of the input item.<br />Available options:  <code class=\\"schema-inline-code\\">input_video</code>"
                      }
                    ]
                  }
                ]
              }
            ]
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the custom tool call output.<br />Available options: <code class=\\"schema-inline-code\\">custom_tool_call_output</code>"
          }
        ]
      }
    ]
  },
  {
    "name": "output_text",
    "type": "string",
    "isBold": true,
    "description": "SDK-only convenience property that contains the aggregated text output from all <code class=\\"schema-inline-code\\">output_text</code> items in the <code class=\\"schema-inline-code\\">output</code> array, if any are present."
  },
  {
    "name": "status",
    "type": "string",
    "isBold": true,
    "description": "The status of the response.<br />Available options: <code class=\\"schema-inline-code\\">completed</code>, <code class=\\"schema-inline-code\\">in_progress</code>, <code class=\\"schema-inline-code\\">incomplete</code>"
  },
  {
    "name": "usage",
    "type": "object",
    "isBold": true,
    "description": "Usage statistics for the response.",
    "children": [
      {
        "name": "ResponseUsage",
        "type": "object",
        "isBold": false,
        "children": [
          {
            "name": "input_tokens",
            "type": "integer",
            "isBold": true,
            "description": "The number of input tokens."
          },
          {
            "name": "input_tokens_details",
            "type": "object",
            "isBold": true,
            "description": "Details about input tokens.",
            "children": [
              {
                "name": "cached_tokens",
                "type": "integer",
                "isBold": true,
                "description": "The number of cached input tokens."
              }
            ]
          },
          {
            "name": "output_tokens",
            "type": "integer",
            "isBold": true,
            "description": "The number of output tokens."
          },
          {
            "name": "output_tokens_details",
            "type": "object",
            "isBold": true,
            "description": "Details about output tokens.",
            "children": [
              {
                "name": "reasoning_tokens",
                "type": "integer",
                "isBold": true,
                "description": "The number of reasoning tokens."
              }
            ]
          },
          {
            "name": "total_tokens",
            "type": "integer",
            "isBold": true,
            "description": "The total number of tokens."
          }
        ]
      }
    ]
  }
]`} />

## Response chunk object (streaming output) 

{/* feishu-style:text-align:left */}
When you create a Response with `stream` set to `true`, the server will emit server-sent events to the client as the Response is generated.

### response.created

> An event that is emitted when a response is created.

<InlineSchemaV2 schema={`[
  {
    "name": "response",
    "type": "object",
    "isBold": true,
    "description": "The response that was created. The parameters contained in this object are identical to those returned by the model creation request in non-streaming mode."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number for this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.created</code>."
  }
]`} />

### response.in_progress

> Emitted when the response is in progress.

<InlineSchemaV2 schema={`[
  {
    "name": "response",
    "type": "object",
    "isBold": true,
    "description": "The response that is in progress. The parameters contained in this object are identical to those returned by the model creation request in non-streaming mode."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.in_progress</code>."
  }
]`} />

### response.completed

> Emitted when the model response is complete.

<InlineSchemaV2 schema={`[
  {
    "name": "response",
    "type": "object",
    "isBold": true,
    "description": "Properties of the completed response. The parameters contained in this object are identical to those returned by the model creation request in non-streaming mode."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number for this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.completed</code>."
  }
]`} />

### response.incomplete

> An event that is emitted when a response finishes as incomplete.

<InlineSchemaV2 schema={`[
  {
    "name": "response",
    "type": "object",
    "isBold": true,
    "description": "The response that was incomplete. The parameters contained in this object are identical to those returned by the model creation request in non-streaming mode."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.incomplete</code>."
  }
]`} />

### response.output_item.added

> Emitted when a new output item is added.

<InlineSchemaV2 schema={`[
  {
    "name": "item",
    "type": "object",
    "isBold": true,
    "description": "The output item that was added. The parameters contained in this object are identical to those of the <code class=\\"schema-inline-code\\">output</code> field returned by the model creation request in non-streaming mode."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item that was added."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.output_item.added</code>."
  }
]`} />

### response.output_item.done

> Emitted when an output item is marked done. 

<InlineSchemaV2 schema={`[
  {
    "name": "item",
    "type": "object",
    "isBold": true,
    "description": "The output item that was marked done. The parameters contained in this object are identical to those of the <code class=\\"schema-inline-code\\">output</code> field returned by the model creation request in non-streaming mode."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item that was marked done."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.output_item.done</code>."
  }
]`} />

### response.content_part.added

> Emitted when a new content part is added.

<InlineSchemaV2 schema={`[
  {
    "name": "content_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the content part that was added."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "The ID of the output item that the content part was added to."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item that the content part was added to."
  },
  {
    "name": "part",
    "type": "object",
    "isBold": true,
    "description": "The content part that was added.",
    "children": [
      {
        "name": "ResponseOutputText",
        "type": "object",
        "isBold": false,
        "description": "A text output from the model.",
        "children": [
          {
            "name": "text",
            "type": "string",
            "isBold": true,
            "description": "The text output from the model."
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the output text. Always <code class=\\"schema-inline-code\\">output_text</code>."
          }
        ]
      },
      {
        "name": "ReasoningText",
        "type": "object",
        "isBold": false,
        "description": "Reasoning text from the model.",
        "children": [
          {
            "name": "text",
            "type": "string",
            "isBold": true,
            "description": "The reasoning text from the model."
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the reasoning text. Always <code class=\\"schema-inline-code\\">reasoning_text</code>."
          }
        ]
      }
    ]
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.content_part.added</code>."
  }
]`} />

### response.content_part.done

> Emitted when a content part is done.

<InlineSchemaV2 schema={`[
  {
    "name": "content_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the content part that is done."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "The ID of the output item that the content part was added to."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item that the content part was added to."
  },
  {
    "name": "part",
    "type": "object",
    "isBold": true,
    "description": "The content part that is done.",
    "children": [
      {
        "name": "ResponseOutputText",
        "type": "object",
        "isBold": false,
        "description": "A text output from the model.",
        "children": [
          {
            "name": "text",
            "type": "string",
            "isBold": true,
            "description": "The text output from the model."
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the output text. Always <code class=\\"schema-inline-code\\">output_text</code>."
          }
        ]
      }
    ]
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.content_part.done</code>."
  }
]`} />

### response.output_text.delta

> Emitted when there is an additional text delta.

<InlineSchemaV2 schema={`[
  {
    "name": "content_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the content part that the text delta was added to."
  },
  {
    "name": "delta",
    "type": "string",
    "isBold": true,
    "description": "The text delta that was added."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "The ID of the output item that the text delta was added to."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item that the text delta was added to."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number for this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.output_text.delta</code>."
  }
]`} />

### response.output_text.done

> Emitted when text content is finalized.

<InlineSchemaV2 schema={`[
  {
    "name": "content_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the content part that the text content is finalized."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "The ID of the output item that the text content is finalized."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item that the text content is finalized."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number for this event."
  },
  {
    "name": "text",
    "type": "string",
    "isBold": true,
    "description": "The text content that is finalized."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.output_text.done</code>."
  }
]`} />

### response.function_call_arguments.delta

> Emitted when there is a partial function-call arguments delta.

<InlineSchemaV2 schema={`[
  {
    "name": "delta",
    "type": "string",
    "isBold": true,
    "description": "The function-call arguments delta that is added."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "The ID of the output item that the function-call arguments delta is added to."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item that the function-call arguments delta is added to."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.function_call_arguments.delta</code>."
  }
]`} />

### response.function_call_arguments.done

> Emitted when function-call arguments are finalized.

<InlineSchemaV2 schema={`[
  {
    "name": "arguments",
    "type": "string",
    "isBold": true,
    "description": "The function-call arguments."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "The ID of the item."
  },
  {
    "name": "name",
    "type": "string",
    "isBold": true,
    "description": "The name of the function that was called."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.function_call_arguments.done</code>."
  }
]`} />

### response.reasoning_text.delta

> Emitted when a delta is added to a reasoning text.

<InlineSchemaV2 schema={`[
  {
    "name": "content_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the reasoning content part this delta is associated with."
  },
  {
    "name": "delta",
    "type": "string",
    "isBold": true,
    "description": "The text delta that was added to the reasoning content."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "The ID of the item this reasoning text delta is associated with."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item this reasoning text delta is associated with."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.reasoning_text.delta</code>."
  }
]`} />

### response.reasoning_text.done

> Emitted when a reasoning text is completed.

<InlineSchemaV2 schema={`[
  {
    "name": "content_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the reasoning content part."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "The ID of the item this reasoning text is associated with."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output item this reasoning text is associated with."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "text",
    "type": "string",
    "isBold": true,
    "description": "The full text of the completed reasoning content."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The type of the event. Always <code class=\\"schema-inline-code\\">response.reasoning_text.done</code>."
  }
]`} />

### response.custom_tool_call_input.delta

> Event representing a delta (partial update) to the input of a custom tool call.

<InlineSchemaV2 schema={`[
  {
    "name": "delta",
    "type": "string",
    "isBold": true,
    "description": "The incremental input data (delta) for the custom tool call."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "Unique identifier for the API item associated with this event."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output this delta applies to."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The event type identifier. Always <code class=\\"schema-inline-code\\">response.custom_tool_call_input.delta</code>."
  }
]`} />

### response.custom_tool_call_input.done

> Event indicating that input for a custom tool call is complete.

<InlineSchemaV2 schema={`[
  {
    "name": "input",
    "type": "string",
    "isBold": true,
    "description": "The complete input data for the custom tool call."
  },
  {
    "name": "item_id",
    "type": "string",
    "isBold": true,
    "description": "Unique identifier for the API item associated with this event."
  },
  {
    "name": "output_index",
    "type": "number",
    "isBold": true,
    "description": "The index of the output this event applies to."
  },
  {
    "name": "sequence_number",
    "type": "number",
    "isBold": true,
    "description": "The sequence number of this event."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "The event type identifier. Always <code class=\\"schema-inline-code\\">response.custom_tool_call_input.done</code>."
  }
]`} />


--- DOCUMENT: Anthropic API ---
URL: https://mimo.mi.com/static/docs/api/chat/anthropic-api.md

# Anthropic Messages API Compatibility

## Request Address

```bash
https://api.xiaomimimo.com/anthropic/v1/messages
```

## Request Headers

{/* feishu-style:text-align:left */}
The API supports the following two authentication methods. Please choose one and add it to the request headers:

<Tab>
  <TabItem label={`API Key Authentication`}>

```json
api-key: $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
  <TabItem label={`Bearer Authentication`}>

```json
Authorization: Bearer $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
</Tab>

## Request Body

<InlineSchemaV2 schema={`[
  {
    "name": "messages",
    "type": "array",
    "isBold": true,
    "required": true,
    "description": "Input messages. Each input message must be an object with a <code class=\\"schema-inline-code\\">role</code> and <code class=\\"schema-inline-code\\">content</code>.<br />Each input message <code class=\\"schema-inline-code\\">content</code> may be either a single <code class=\\"schema-inline-code\\">string</code> or an array of content blocks, where each block has a specific <code class=\\"schema-inline-code\\">type</code>. Using a <code class=\\"schema-inline-code\\">string</code> for <code class=\\"schema-inline-code\\">content</code> is shorthand for an array of one content block of type <code class=\\"schema-inline-code\\">text</code>.",
    "children": [
      {
        "name": "role",
        "type": "string",
        "isBold": true,
        "required": true,
        "description": "Role of the message.<br />Available options: <code class=\\"schema-inline-code\\">user</code>, <code class=\\"schema-inline-code\\">assistant</code>, <code class=\\"schema-inline-code\\">system</code>"
      },
      {
        "name": "content",
        "type": [
          "string",
          "array"
        ],
        "isBold": true,
        "required": true,
        "children": [
          {
            "name": "Text content",
            "type": "string",
            "isBold": false,
            "description": "The text contents of the message."
          },
          {
            "name": "Array of content parts",
            "type": "array",
            "isBold": false,
            "description": "An array of content parts with a defined type. Such text, image, audio, video, tool use, tool result, and thinking.<br /><blockquote class=\\"schema-blockquote\\">Currently, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models support image, audio or video input.</blockquote>",
            "children": [
              {
                "name": "Text",
                "type": "object",
                "isBold": false,
                "children": [
                  {
                    "name": "text",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The content of the text block.<br />Minimum length: <code class=\\"schema-inline-code\\">1</code>"
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
                  }
                ]
              },
              {
                "name": "Image",
                "type": "object",
                "isBold": false,
                "children": [
                  {
                    "name": "source",
                    "type": "object",
                    "isBold": true,
                    "required": true,
                    "description": "Image data is provided via URL or Base64.",
                    "children": [
                      {
                        "name": "Base64ImageSource",
                        "type": "object",
                        "isBold": false,
                        "children": [
                          {
                            "name": "data",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Base64 encoded image data."
                          },
                          {
                            "name": "media_type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Media type.<br />Available options: <code class=\\"schema-inline-code\\">image/jpeg</code>, <code class=\\"schema-inline-code\\">image/png</code>, <code class=\\"schema-inline-code\\">image/gif</code>, <code class=\\"schema-inline-code\\">image/webp</code>, <code class=\\"schema-inline-code\\">image/bmp</code>"
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Image source type.<br />Available options: <code class=\\"schema-inline-code\\">base64</code>"
                          }
                        ]
                      },
                      {
                        "name": "URLImageSource",
                        "type": "object",
                        "isBold": false,
                        "children": [
                          {
                            "name": "url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "A URL of the image."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Image source type.<br />Available options: <code class=\\"schema-inline-code\\">url</code>"
                          }
                        ]
                      }
                    ]
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">image</code>"
                  }
                ]
              },
              {
                "name": "Audio",
                "type": "object",
                "isBold": false,
                "children": [
                  {
                    "name": "source",
                    "type": "object",
                    "isBold": true,
                    "required": true,
                    "description": "Audio data is provided via URL or Base64.",
                    "children": [
                      {
                        "name": "Base64AudioSource",
                        "type": "object",
                        "isBold": false,
                        "children": [
                          {
                            "name": "data",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Base64 encoded audio data."
                          },
                          {
                            "name": "media_type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Media type.<br />Available options: <code class=\\"schema-inline-code\\">audio/mpeg</code>, <code class=\\"schema-inline-code\\">audio/wav</code>, <code class=\\"schema-inline-code\\">audio/flac</code>, <code class=\\"schema-inline-code\\">audio/mp4</code>, <code class=\\"schema-inline-code\\">audio/ogg</code>"
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Audio source type.<br />Available options: <code class=\\"schema-inline-code\\">base64</code>"
                          }
                        ]
                      },
                      {
                        "name": "URLAudioSource",
                        "type": "object",
                        "isBold": false,
                        "children": [
                          {
                            "name": "url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "A URL of the audio."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Audio source type.<br />Available options: <code class=\\"schema-inline-code\\">url</code>"
                          }
                        ]
                      }
                    ]
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">audio</code>"
                  }
                ]
              },
              {
                "name": "Video",
                "type": "object",
                "isBold": false,
                "children": [
                  {
                    "name": "source",
                    "type": "object",
                    "isBold": true,
                    "required": true,
                    "description": "Video data is provided via URL or Base64.",
                    "children": [
                      {
                        "name": "Base64VideoSource",
                        "type": "object",
                        "isBold": false,
                        "children": [
                          {
                            "name": "data",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Base64 encoded video data."
                          },
                          {
                            "name": "media_type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Media type.<br />Available options: <code class=\\"schema-inline-code\\">video/mp4</code>, <code class=\\"schema-inline-code\\">video/quicktime</code>, <code class=\\"schema-inline-code\\">video/x-msvideo</code>, <code class=\\"schema-inline-code\\">video/x-ms-wmv</code>"
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Video source type.<br />Available options: <code class=\\"schema-inline-code\\">base64</code>"
                          }
                        ]
                      },
                      {
                        "name": "URLVideoSource",
                        "type": "object",
                        "isBold": false,
                        "children": [
                          {
                            "name": "url",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "A URL of the video."
                          },
                          {
                            "name": "type",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Video source type.<br />Available options: <code class=\\"schema-inline-code\\">url</code>"
                          }
                        ]
                      },
                      {
                        "name": "fps",
                        "type": "number",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "2",
                        "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                      },
                      {
                        "name": "media_resolution",
                        "type": "string",
                        "isBold": true,
                        "required": false,
                        "defaultValue": "default",
                        "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                      }
                    ]
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">video</code>"
                  }
                ]
              },
              {
                "name": "Tool use",
                "type": "object",
                "isBold": false,
                "children": [
                  {
                    "name": "id",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The unique identifier for tool use."
                  },
                  {
                    "name": "input",
                    "type": "object",
                    "isBold": true,
                    "required": true,
                    "description": "The parameter object passed when using the tool."
                  },
                  {
                    "name": "name",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "Tool name."
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">tool_use</code>"
                  }
                ]
              },
              {
                "name": "Tool result",
                "type": "object",
                "isBold": false,
                "children": [
                  {
                    "name": "tool_use_id",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The <code class=\\"schema-inline-code\\">tool_use</code> ID corresponding to this result."
                  },
                  {
                    "name": "content",
                    "type": [
                      "string",
                      "array"
                    ],
                    "isBold": true,
                    "description": "The result returned after the tool is executed.",
                    "children": [
                      {
                        "name": "Text content",
                        "type": "string",
                        "isBold": false,
                        "description": "The text contents of the message."
                      },
                      {
                        "name": "Array of content parts",
                        "type": "array",
                        "isBold": false,
                        "description": "An array of content parts with a defined type. Such text, image, audio and video.",
                        "children": [
                          {
                            "name": "Text",
                            "type": "object",
                            "isBold": false,
                            "children": [
                              {
                                "name": "text",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "The content of the text block."
                              },
                              {
                                "name": "type",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
                              }
                            ]
                          },
                          {
                            "name": "Image",
                            "type": "object",
                            "isBold": false,
                            "children": [
                              {
                                "name": "source",
                                "type": "object",
                                "isBold": true,
                                "required": true,
                                "description": "Image data is provided via URL or Base64.",
                                "children": [
                                  {
                                    "name": "Base64ImageSource",
                                    "type": "object",
                                    "isBold": false,
                                    "children": [
                                      {
                                        "name": "data",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Base64 encoded image data."
                                      },
                                      {
                                        "name": "media_type",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Media type.<br />Available options: <code class=\\"schema-inline-code\\">image/jpeg</code>, <code class=\\"schema-inline-code\\">image/png</code>, <code class=\\"schema-inline-code\\">image/gif</code>, <code class=\\"schema-inline-code\\">image/webp</code>, <code class=\\"schema-inline-code\\">image/bmp</code>"
                                      },
                                      {
                                        "name": "type",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Image source type.<br />Available options: <code class=\\"schema-inline-code\\">base64</code>"
                                      }
                                    ]
                                  },
                                  {
                                    "name": "URLImageSource",
                                    "type": "object",
                                    "isBold": false,
                                    "children": [
                                      {
                                        "name": "url",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "A URL of the image."
                                      },
                                      {
                                        "name": "type",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Image source type.<br />Available options: <code class=\\"schema-inline-code\\">url</code>"
                                      }
                                    ]
                                  }
                                ]
                              },
                              {
                                "name": "type",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">image</code>"
                              }
                            ]
                          },
                          {
                            "name": "Audio",
                            "type": "object",
                            "isBold": false,
                            "children": [
                              {
                                "name": "source",
                                "type": "object",
                                "isBold": true,
                                "required": true,
                                "description": "Audio data is provided via URL or Base64.",
                                "children": [
                                  {
                                    "name": "Base64AudioSource",
                                    "type": "object",
                                    "isBold": false,
                                    "children": [
                                      {
                                        "name": "data",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Base64 encoded audio data."
                                      },
                                      {
                                        "name": "media_type",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Media type.<br />Available options: <code class=\\"schema-inline-code\\">audio/mpeg</code>, <code class=\\"schema-inline-code\\">audio/wav</code>, <code class=\\"schema-inline-code\\">audio/flac</code>, <code class=\\"schema-inline-code\\">audio/mp4</code>, <code class=\\"schema-inline-code\\">audio/ogg</code>"
                                      },
                                      {
                                        "name": "type",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Audio source type.<br />Available options: <code class=\\"schema-inline-code\\">base64</code>"
                                      }
                                    ]
                                  },
                                  {
                                    "name": "URLAudioSource",
                                    "type": "object",
                                    "isBold": false,
                                    "children": [
                                      {
                                        "name": "url",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "A URL of the audio."
                                      },
                                      {
                                        "name": "type",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Audio source type.<br />Available options: <code class=\\"schema-inline-code\\">url</code>"
                                      }
                                    ]
                                  }
                                ]
                              },
                              {
                                "name": "type",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">audio</code>"
                              }
                            ]
                          },
                          {
                            "name": "Video",
                            "type": "object",
                            "isBold": false,
                            "children": [
                              {
                                "name": "source",
                                "type": "object",
                                "isBold": true,
                                "required": true,
                                "description": "Video data is provided via URL or Base64.",
                                "children": [
                                  {
                                    "name": "Base64VideoSource",
                                    "type": "object",
                                    "isBold": false,
                                    "children": [
                                      {
                                        "name": "data",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Base64 encoded video data."
                                      },
                                      {
                                        "name": "media_type",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Media type.<br />Available options: <code class=\\"schema-inline-code\\">video/mp4</code>, <code class=\\"schema-inline-code\\">video/quicktime</code>, <code class=\\"schema-inline-code\\">video/x-msvideo</code>, <code class=\\"schema-inline-code\\">video/x-ms-wmv</code>"
                                      },
                                      {
                                        "name": "type",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Video source type.<br />Available options: <code class=\\"schema-inline-code\\">base64</code>"
                                      }
                                    ]
                                  },
                                  {
                                    "name": "URLVideoSource",
                                    "type": "object",
                                    "isBold": false,
                                    "children": [
                                      {
                                        "name": "url",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "A URL of the video."
                                      },
                                      {
                                        "name": "type",
                                        "type": "string",
                                        "isBold": true,
                                        "required": true,
                                        "description": "Video source type.<br />Available options: <code class=\\"schema-inline-code\\">url</code>"
                                      }
                                    ]
                                  },
                                  {
                                    "name": "fps",
                                    "type": "number",
                                    "isBold": true,
                                    "required": false,
                                    "defaultValue": "2",
                                    "description": "Number of frames sampled per second.<br />Required range: <code class=\\"schema-inline-code\\">[0.1, 10.0]</code>"
                                  },
                                  {
                                    "name": "media_resolution",
                                    "type": "string",
                                    "isBold": true,
                                    "required": false,
                                    "defaultValue": "default",
                                    "description": "Resolution level.<br />Available options: <code class=\\"schema-inline-code\\">default</code>, <code class=\\"schema-inline-code\\">max</code>"
                                  }
                                ]
                              },
                              {
                                "name": "type",
                                "type": "string",
                                "isBold": true,
                                "required": true,
                                "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">video</code>"
                              }
                            ]
                          }
                        ]
                      }
                    ]
                  },
                  {
                    "name": "is_error",
                    "type": "boolean",
                    "isBold": true
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">tool_result</code>"
                  }
                ]
              },
              {
                "name": "Thinking",
                "type": "object",
                "isBold": false,
                "children": [
                  {
                    "name": "signature",
                    "type": "string",
                    "isBold": true,
                    "description": "The signature of the thinking block."
                  },
                  {
                    "name": "thinking",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "Thinking content."
                  },
                  {
                    "name": "type",
                    "type": "string",
                    "isBold": true,
                    "required": true,
                    "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">thinking</code>"
                  }
                ]
              }
            ]
          }
        ]
      }
    ]
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "required": true,
    "description": "The model that will complete your prompt.<br />Available options: <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code>"
  },
  {
    "name": "max_tokens",
    "type": "integer",
    "isBold": true,
    "required": false,
    "description": "The maximum number of tokens to generate before stopping.<br />Note that our models may stop before reaching this maximum. This parameter only specifies the absolute maximum number of tokens to generate.<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>: default <code class=\\"schema-inline-code\\">131072</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5</code>: default <code class=\\"schema-inline-code\\">32768</code></li></ul>Required range: <code class=\\"schema-inline-code\\">[1, 131072]</code>"
  },
  {
    "name": "stop_sequences",
    "type": "array",
    "isBold": true,
    "required": false,
    "description": "Custom text sequences that will cause the model to stop generating.<br />Our models will normally stop when they have naturally completed their turn, which will result in a response <code class=\\"schema-inline-code\\">stop_reason</code> of <code class=\\"schema-inline-code\\">end_turn</code>.<br />If you want the model to stop generating when it encounters custom strings of text, you can use the <code class=\\"schema-inline-code\\">stop_sequences</code> parameter."
  },
  {
    "name": "stream",
    "type": "boolean",
    "isBold": true,
    "required": false,
    "defaultValue": "false",
    "description": "Whether to incrementally stream the response using server-sent events."
  },
  {
    "name": "system",
    "type": [
      "string",
      "array"
    ],
    "isBold": true,
    "required": false,
    "description": "A system prompt is a way of providing context and instructions to model, such as specifying a particular goal or role.",
    "children": [
      {
        "name": "Text content",
        "type": "string",
        "isBold": false,
        "description": "The content of the system prompt."
      },
      {
        "name": "Array of content parts",
        "type": "array",
        "isBold": false,
        "children": [
          {
            "name": "text",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The text content.<br />Minimum length: <code class=\\"schema-inline-code\\">1</code>"
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
          }
        ]
      }
    ]
  },
  {
    "name": "temperature",
    "type": "number",
    "isBold": true,
    "required": false,
    "description": "Sampling temperature controls the diversity of the text generated by the model.<br />The higher the temperature, the more diverse the generated text will be; conversely, the lower the temperature, the more deterministic the generated text will be.<br /><blockquote class=\\"schema-blockquote\\">In thinking mode, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models do not support customizing the <code class=\\"schema-inline-code\\">temperature</code> parameter. Even if this parameter is passed in, it will be forcibly overridden and take effect with the model's recommended default value of <code class=\\"schema-inline-code\\">1.0</code>.</blockquote><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code>: default <code class=\\"schema-inline-code\\">1.0</code></li></ul>Required range: <code class=\\"schema-inline-code\\">[0, 1.5]</code>"
  },
  {
    "name": "thinking",
    "type": "object",
    "isBold": true,
    "required": false,
    "description": "Configuration for enabling model's extended thinking.<br /><blockquote class=\\"schema-blockquote\\">Note: During the multi-turn tool calls process in thinking mode, the model returns a <code class=\\"schema-inline-code\\">thinking</code> content block alongside <code class=\\"schema-inline-code\\">tool_use</code> content block. To continue the conversation, it is recommended to keep all previous <code class=\\"schema-inline-code\\">thinking</code> content block in the <code class=\\"schema-inline-code\\">messages</code> array for each subsequent request to achieve the best performance.</blockquote><blockquote class=\\"schema-blockquote\\">In thinking mode, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models do not support customizing the <code class=\\"schema-inline-code\\">temperature</code> and <code class=\\"schema-inline-code\\">top_p</code> parameters. Even if these parameters are passed in, the actual effective values will be forcibly set by the model to its recommended default values of <code class=\\"schema-inline-code\\">1.0</code> and <code class=\\"schema-inline-code\\">0.95</code>.</blockquote>",
    "children": [
      {
        "name": "type",
        "type": "string",
        "isBold": true,
        "required": true,
        "description": "<ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.5</code>: default <code class=\\"schema-inline-code\\">enabled</code></li></ul>Available options: <code class=\\"schema-inline-code\\">enabled</code>, <code class=\\"schema-inline-code\\">disabled</code>"
      }
    ]
  },
  {
    "name": "tool_choice",
    "type": "object",
    "isBold": true,
    "required": false,
    "description": "How the model should use the provided tools.",
    "children": [
      {
        "name": "type",
        "type": "string",
        "isBold": true,
        "required": true,
        "description": "<ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">auto</code> means the model will automatically decide whether to use tools.</li></ul><blockquote class=\\"schema-blockquote\\">Note: When a value other than <code class=\\"schema-inline-code\\">auto</code> is passed to <code class=\\"schema-inline-code\\">type</code>, the backend will remove this field by default, and the model response behavior will still be equivalent to the <code class=\\"schema-inline-code\\">auto</code> mode (this logic is subject to future adjustments).</blockquote>Available options: <code class=\\"schema-inline-code\\">auto</code>"
      },
      {
        "name": "disable_parallel_tool_use",
        "type": "boolean",
        "isBold": true,
        "defaultValue": "false",
        "description": "Whether to disable parallel tool use.<br />If set to <code class=\\"schema-inline-code\\">true</code>:<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\">When type is <code class=\\"schema-inline-code\\">auto</code>, the model will output at most one tool use.</li></ul>"
      }
    ]
  },
  {
    "name": "tools",
    "type": "array",
    "isBold": true,
    "required": false,
    "description": "Definitions of tools that the model may use.<br />If you include <code class=\\"schema-inline-code\\">tools</code> in your API request, the model may return <code class=\\"schema-inline-code\\">tool_use</code> content blocks that represent the model's use of those tools. You can then run those tools using the tool input generated by the model and then optionally return results back to the model using <code class=\\"schema-inline-code\\">tool_result</code> content blocks.<br /><blockquote class=\\"schema-blockquote\\">Note: During the multi-turn tool calls process in thinking mode, the model returns a <code class=\\"schema-inline-code\\">thinking</code> content block alongside <code class=\\"schema-inline-code\\">tool_use</code> content block. To continue the conversation, it is recommended to keep all previous <code class=\\"schema-inline-code\\">thinking</code> content block in the <code class=\\"schema-inline-code\\">messages</code> array for each subsequent request to achieve the best performance.</blockquote>Each tool definition includes:<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">name</code>: Name of the tool.</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">description</code>: Optional, but strongly-recommended description of the tool.</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">input_schema</code>:  JSON schema for the tool <code class=\\"schema-inline-code\\">input</code> shape that the model will produce in <code class=\\"schema-inline-code\\">tool_use</code> output content blocks.</li></ul>",
    "children": [
      {
        "name": "name",
        "type": "string",
        "isBold": true,
        "required": true,
        "description": "Name of the tool.<br />This is how the tool will be called by the model and in <code class=\\"schema-inline-code\\">tool_use</code> blocks."
      },
      {
        "name": "description",
        "type": "string",
        "isBold": true,
        "description": "Description of what this tool does.<br />Tool descriptions should be as detailed as possible. The more information that the model has about what the tool is and how to use it, the better it will perform. You can use natural language descriptions to reinforce important aspects of the tool input JSON schema."
      },
      {
        "name": "type",
        "type": "string",
        "isBold": true,
        "description": "Available options: <code class=\\"schema-inline-code\\">custom</code>"
      },
      {
        "name": "input_schema",
        "type": "object",
        "isBold": true,
        "required": true,
        "description": "JSON schema for the tool input shape that the model will produce in <code class=\\"schema-inline-code\\">tool_use</code> output content blocks.",
        "children": [
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The type of <code class=\\"schema-inline-code\\">input_schema</code>, only <code class=\\"schema-inline-code\\">object</code> is supported.<br />Available options: <code class=\\"schema-inline-code\\">object</code>"
          },
          {
            "name": "properties",
            "type": [
              "object",
              "null"
            ],
            "isBold": true,
            "description": "The properties of the tool input."
          },
          {
            "name": "required",
            "type": [
              "array",
              "null"
            ],
            "isBold": true,
            "description": "The list of properties that must be included in the tool input."
          }
        ]
      }
    ]
  },
  {
    "name": "top_p",
    "type": "number",
    "isBold": true,
    "required": false,
    "defaultValue": "0.95",
    "description": "Use nucleus sampling.<br />In nucleus sampling, we compute the cumulative distribution over all the options for each subsequent token in decreasing probability order and cut it off once it reaches a particular probability specified by <code class=\\"schema-inline-code\\">top_p</code>. You should either alter <code class=\\"schema-inline-code\\">temperature</code> or <code class=\\"schema-inline-code\\">top_p</code>, but not both.<br />Recommended for advanced use cases only. You usually only need to use <code class=\\"schema-inline-code\\">temperature</code>.<br /><blockquote class=\\"schema-blockquote\\">In thinking mode, the <code class=\\"schema-inline-code\\">mimo-v2.6-flash</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro</code>, <code class=\\"schema-inline-code\\">mimo-v2.6-pro-ultraspeed</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-pro</code> and <code class=\\"schema-inline-code\\">mimo-v2.5</code> models do not support customizing the <code class=\\"schema-inline-code\\">top_p</code> parameter. Even if this parameter is passed in, it will be forcibly overridden and take effect with the model's recommended default value of <code class=\\"schema-inline-code\\">0.95</code>.</blockquote>Required range: <code class=\\"schema-inline-code\\">[0.01, 1.0]</code>"
  }
]`} />

## Non-streaming Response

<InlineSchemaV2 schema={`[
  {
    "name": "id",
    "type": "string",
    "isBold": true,
    "description": "Unique object identifier. The format and length of IDs may change over time."
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "For Messages, this is always <code class=\\"schema-inline-code\\">message</code>."
  },
  {
    "name": "role",
    "type": "string",
    "isBold": true,
    "description": "Conversational role of the generated message. This will always be <code class=\\"schema-inline-code\\">assistant</code>."
  },
  {
    "name": "content",
    "type": "array",
    "isBold": true,
    "description": "Content generated by the model.",
    "children": [
      {
        "name": "Text",
        "type": "object",
        "isBold": false,
        "children": [
          {
            "name": "text",
            "type": "string",
            "isBold": true,
            "description": "The content of the text."
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
          }
        ]
      },
      {
        "name": "Thinking",
        "type": "object",
        "isBold": false,
        "children": [
          {
            "name": "signature",
            "type": "string",
            "isBold": true,
            "description": "The signature of the thinking block."
          },
          {
            "name": "thinking",
            "type": "string",
            "isBold": true,
            "description": "Thinking content."
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">thinking</code>"
          }
        ]
      },
      {
        "name": "Tool use",
        "type": "object",
        "isBold": false,
        "children": [
          {
            "name": "id",
            "type": "string",
            "isBold": true,
            "description": "The unique identifier for tool use."
          },
          {
            "name": "input",
            "type": "object",
            "isBold": true,
            "description": "The parameter object passed when using the tool."
          },
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "description": "Tool name."
          },
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The type of the content.<br />Available options: <code class=\\"schema-inline-code\\">tool_use</code>"
          }
        ]
      }
    ]
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "description": "The model that handled the request."
  },
  {
    "name": "stop_reason",
    "type": "string",
    "isBold": true,
    "description": "The reason the message finished.<br />This may be one the following values:<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">end_turn</code>: the model reached a natural stopping point.</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">max_tokens</code>: we exceeded the requested <code class=\\"schema-inline-code\\">max_tokens</code> or the model's maximum.</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">tool_use</code>: the model invoked one or more tools.</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">content_filter</code>: the content was omitted due to a flag from our content filters.</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">repetition_truncation</code>: the model detects repetition.</li></ul>Available options: <code class=\\"schema-inline-code\\">end_turn</code>, <code class=\\"schema-inline-code\\">max_tokens</code>, <code class=\\"schema-inline-code\\">tool_use</code>, <code class=\\"schema-inline-code\\">content_filter</code>, <code class=\\"schema-inline-code\\">repetition_truncation</code>"
  },
  {
    "name": "usage",
    "type": "object",
    "isBold": true,
    "description": "Billing and rate-limit usage.",
    "children": [
      {
        "name": "input_tokens",
        "type": "integer",
        "isBold": true,
        "description": "The number of input tokens which were used."
      },
      {
        "name": "output_tokens",
        "type": "integer",
        "isBold": true,
        "description": "The number of output tokens which were used."
      },
      {
        "name": "cache_read_input_tokens",
        "type": [
          "integer",
          "null"
        ],
        "isBold": true,
        "description": "The number of input tokens read from the cache."
      }
    ]
  }
]`} />

## Streaming Response
<InlineSchemaV2 schema={`[
  {
    "name": "SSE.event",
    "type": "string",
    "isBold": true,
    "description": "A string identifying the type of event described.<br />Available options: <code class=\\"schema-inline-code\\">message_start</code>, <code class=\\"schema-inline-code\\">content_block_start</code>,<code class=\\"schema-inline-code\\"> content_block_delta</code>, <code class=\\"schema-inline-code\\">content_block_stop</code>, <code class=\\"schema-inline-code\\">message_delta</code>, <code class=\\"schema-inline-code\\">message_stop</code>"
  },
  {
    "name": "type",
    "type": "string",
    "isBold": true,
    "description": "Each server-sent event includes a named event type and associated JSON data.<br />Available options: <code class=\\"schema-inline-code\\">message_start</code>, <code class=\\"schema-inline-code\\">content_block_start</code>,<code class=\\"schema-inline-code\\"> content_block_delta</code>, <code class=\\"schema-inline-code\\">content_block_stop</code>, <code class=\\"schema-inline-code\\">message_delta</code>, <code class=\\"schema-inline-code\\">message_stop</code>"
  },
  {
    "name": "message",
    "type": "object",
    "isBold": true,
    "description": "Response message.",
    "children": [
      {
        "name": "id",
        "type": "string",
        "isBold": true,
        "description": "The message ID."
      },
      {
        "name": "type",
        "type": "string",
        "isBold": true,
        "description": "Available options: <code class=\\"schema-inline-code\\">message</code>"
      },
      {
        "name": "role",
        "type": "string",
        "isBold": true,
        "description": "Available options: <code class=\\"schema-inline-code\\">assistant</code>"
      },
      {
        "name": "model",
        "type": "string",
        "isBold": true,
        "description": "The model name."
      },
      {
        "name": "content",
        "type": "array",
        "isBold": true,
        "description": "The array of content blocks in the message."
      },
      {
        "name": "stop_reason",
        "type": [
          "string",
          "null"
        ],
        "isBold": true,
        "description": "The reason the message finished."
      }
    ]
  },
  {
    "name": "index",
    "type": "integer",
    "isBold": true,
    "description": "The position of the content block within the message.。"
  },
  {
    "name": "content_block",
    "type": "object",
    "isBold": true,
    "description": "The content block that is starting.",
    "children": [
      {
        "name": "Text",
        "type": "object",
        "isBold": false,
        "children": [
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The header for a text content block; actual text arrives via subsequent delta events.<br />Available options: <code class=\\"schema-inline-code\\">text</code>"
          },
          {
            "name": "text",
            "type": "string",
            "isBold": true,
            "description": "Often an empty string at start; text is appended via <code class=\\"schema-inline-code\\">content_block_delta</code> events of type <code class=\\"schema-inline-code\\">text_delta</code>."
          }
        ]
      },
      {
        "name": "Thinking",
        "type": "object",
        "isBold": false,
        "children": [
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "The header for a thinking content block; actual thinking content arrives via subsequent delta events.<br />Available options: <code class=\\"schema-inline-code\\">thinking</code>"
          },
          {
            "name": "thinking",
            "type": "string",
            "isBold": true,
            "description": "Often an empty string at start; thinking content is appended via <code class=\\"schema-inline-code\\">content_block_delta</code> events of type <code class=\\"schema-inline-code\\">thinking_delta</code>."
          }
        ]
      },
      {
        "name": "Tool use",
        "type": "object",
        "isBold": false,
        "children": [
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "Available options: <code class=\\"schema-inline-code\\">tool_use</code>"
          },
          {
            "name": "id",
            "type": "string",
            "isBold": true,
            "description": "The unique identifier for tool use."
          },
          {
            "name": "name",
            "type": "string",
            "isBold": true,
            "description": "Tool name."
          },
          {
            "name": "input",
            "type": "object",
            "isBold": true,
            "description": "The parameter object passed when using the tool."
          }
        ]
      }
    ]
  },
  {
    "name": "delta",
    "type": "object",
    "isBold": true,
    "description": "Actual response content.",
    "children": [
      {
        "name": "Content block delta",
        "type": "object",
        "isBold": false,
        "description": "Incremental data for a content block.",
        "children": [
          {
            "name": "type",
            "type": "string",
            "isBold": true,
            "description": "Available options: <code class=\\"schema-inline-code\\">text_delta</code>, <code class=\\"schema-inline-code\\">thinking_delta</code>, <code class=\\"schema-inline-code\\">input_json_delta</code>"
          },
          {
            "name": "text",
            "type": "string",
            "isBold": true,
            "description": "The text part of the incremental data."
          },
          {
            "name": "thinking",
            "type": "string",
            "isBold": true,
            "description": "The thinking part of the incremental data."
          },
          {
            "name": "partial_json",
            "type": "string",
            "isBold": true,
            "description": "A JSON fragment string. Clients should concatenate fragments in arrival order to form the complete input JSON, then parse."
          }
        ]
      },
      {
        "name": "Message delta",
        "type": "object",
        "isBold": false,
        "description": "Message-level stop metadata updates.",
        "children": [
          {
            "name": "stop_reason",
            "type": [
              "string",
              "null"
            ],
            "isBold": true,
            "description": "The reason the message finished.<br />Available options: <code class=\\"schema-inline-code\\">end_turn</code>, <code class=\\"schema-inline-code\\">max_tokens</code>, <code class=\\"schema-inline-code\\">tool_use</code>, <code class=\\"schema-inline-code\\">content_filter</code>, <code class=\\"schema-inline-code\\">repetition_truncation</code>"
          }
        ]
      }
    ]
  },
  {
    "name": "usage",
    "type": [
      "object",
      "null"
    ],
    "isBold": true,
    "description": "Billing and rate-limit usage.",
    "children": [
      {
        "name": "input_tokens",
        "type": "integer",
        "isBold": true,
        "description": "The number of input tokens which were used."
      },
      {
        "name": "output_tokens",
        "type": "integer",
        "isBold": true,
        "description": "The number of output tokens which were used."
      },
      {
        "name": "cache_read_input_tokens",
        "type": [
          "integer",
          "null"
        ],
        "isBold": true,
        "description": "The number of input tokens read from the cache."
      }
    ]
  }
]`} />


--- DOCUMENT: Speech Recognition (MiMo‑V2.5-ASR) - OpenAI API Compatibility ---
URL: https://mimo.mi.com/static/docs/api/audio/Speech-Recognition.md

# Speech Recognition (MiMo‑V2.5-ASR) - OpenAI API Compatibility

## Request Address

```bash
https://api.xiaomimimo.com/v1/chat/completions
```

## Request Headers

{/* feishu-style:text-align:left */}
The API supports the following two authentication methods. Please choose one and add it to the request headers:

<Tab>
  <TabItem label={`API Key Authentication`}>

```json
api-key: $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
  <TabItem label={`Bearer Authentication`}>

```json
Authorization: Bearer $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
</Tab>

## Request body 

<InlineSchemaV2 schema={`[
  {
    "name": "messages",
    "type": "array",
    "isBold": true,
    "required": true,
    "description": "The message list.",
    "children": [
      {
        "name": "User message",
        "type": "object",
        "isBold": false,
        "description": "Messages sent by an end user.",
        "children": [
          {
            "name": "content",
            "type": "array",
            "isBold": true,
            "required": true,
            "description": "The contents of the user message.<br /><blockquote class=\\"schema-blockquote\\">For detailed usage, please refer to <a target=\\"_blank\\" rel=\\"noopener noreferrer\\" href=\\"https://platform.xiaomimimo.com/docs/en-US/usage-guide/Speech-Recognition\\">Speech Recognition</a>.</blockquote>",
            "children": [
              {
                "name": "Array of content parts",
                "type": "array",
                "isBold": false,
                "description": "An array of content parts with a defined type. For speech recognition, only single audio input is supported.",
                "children": [
                  {
                    "name": "Audio content part",
                    "type": "object",
                    "isBold": false,
                    "children": [
                      {
                        "name": "input_audio",
                        "type": "object",
                        "isBold": true,
                        "required": true,
                        "description": "<blockquote class=\\"schema-blockquote\\">When audio is passed via data URL, the <code class=\\"schema-inline-code\\">format</code> field is optional. If only Base64-encoded audio data is provided, the <code class=\\"schema-inline-code\\">format</code> field is required. If both <code class=\\"schema-inline-code\\">MIME_TYPE</code> and <code class=\\"schema-inline-code\\">format</code> are included, their values must match.</blockquote>",
                        "children": [
                          {
                            "name": "data",
                            "type": "string",
                            "isBold": true,
                            "required": true,
                            "description": "Base64 encoded audio in a data URL. Input audio only supports <code class=\\"schema-inline-code\\">mp3</code> and <code class=\\"schema-inline-code\\">wav</code> formats:<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mp3</code>: valid <code class=\\"schema-inline-code\\">MIME_TYPE</code> values: <code class=\\"schema-inline-code\\">audio/mpeg</code>, <code class=\\"schema-inline-code\\">audio/mp3</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">wav</code>: valid <code class=\\"schema-inline-code\\">MIME_TYPE</code> value: <code class=\\"schema-inline-code\\">audio/wav</code></li></ul>"
                          },
                          {
                            "name": "format",
                            "type": "string",
                            "isBold": true,
                            "description": "The format for encoding audio data.<br />Available options: <code class=\\"schema-inline-code\\">mp3</code>, <code class=\\"schema-inline-code\\">wav</code>"
                          }
                        ]
                      },
                      {
                        "name": "type",
                        "type": "string",
                        "isBold": true,
                        "required": true,
                        "description": "The type of the content part.<br />Available options: <code class=\\"schema-inline-code\\">input_audio</code>"
                      }
                    ]
                  }
                ]
              }
            ]
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Role of the message author.<br />Available options: <code class=\\"schema-inline-code\\">user</code>"
          }
        ]
      }
    ]
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "required": true,
    "description": "Model ID is used to generate the response.<br />Available options: <code class=\\"schema-inline-code\\">mimo-v2.5-asr</code>"
  },
  {
    "name": "asr_options",
    "type": "object",
    "isBold": true,
    "required": false,
    "description": "Custom configuration parameters for automatic speech recognition (ASR).",
    "children": [
      {
        "name": "language",
        "type": "string",
        "isBold": true,
        "required": false,
        "defaultValue": "auto",
        "description": "Specify a single language for audio recognition.<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">auto</code>: Auto‑detect audio language</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">zh</code>: Chinese</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">en</code>: English</li></ul>Available options: <code class=\\"schema-inline-code\\">auto</code>, <code class=\\"schema-inline-code\\">zh</code>, <code class=\\"schema-inline-code\\">en</code>"
      }
    ]
  },
  {
    "name": "stream",
    "type": "boolean",
    "isBold": true,
    "required": false,
    "defaultValue": "false",
    "description": "If set to <code class=\\"schema-inline-code\\">true</code>, the model response data will be streamed to the client as it is generated using server-sent events."
  }
]`} />

## Chat response object (non-streaming output) 

<InlineSchemaV2 schema={`[
  {
    "name": "choices",
    "type": "array",
    "isBold": true,
    "description": "A list of chat completion choices.",
    "children": [
      {
        "name": "finish_reason",
        "type": "string",
        "isBold": true,
        "description": "The reason the model stopped generating tokens:<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">stop</code>: The model reached a natural stop point or a user‑provided stop sequence</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">length</code>: Terminated due to exceeding the model's maximum generation length</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">content_filter</code>: Content was omitted due to a content filter flag</li></ul>"
      },
      {
        "name": "index",
        "type": "integer",
        "isBold": true,
        "description": "The index of the choice in the list of choices."
      },
      {
        "name": "message",
        "type": "object",
        "isBold": true,
        "description": "A chat completion message generated by the model.",
        "children": [
          {
            "name": "content",
            "type": "string",
            "isBold": true,
            "description": "The contents of the message."
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "description": "The role of the author of this message."
          }
        ]
      }
    ]
  },
  {
    "name": "created",
    "type": "integer",
    "isBold": true,
    "description": "The Unix timestamp (in seconds) of when the chat completion was created."
  },
  {
    "name": "id",
    "type": "string",
    "isBold": true,
    "description": "A unique identifier for the chat completion."
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "description": "The model to generate the completion."
  },
  {
    "name": "object",
    "type": "string",
    "isBold": true,
    "description": "The object type, which is always <code class=\\"schema-inline-code\\">chat.completion</code>."
  },
  {
    "name": "usage",
    "type": [
      "object",
      "null"
    ],
    "isBold": true,
    "description": "Usage statistics for the completion request.",
    "children": [
      {
        "name": "completion_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the generated completion."
      },
      {
        "name": "prompt_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the prompt."
      },
      {
        "name": "total_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Total number of tokens used in the request (prompt + completion)."
      },
      {
        "name": "completion_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in a completion.",
        "children": [
          {
            "name": "reasoning_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Tokens generated by the model for reasoning. Always <code class=\\"schema-inline-code\\">0</code>."
          }
        ]
      },
      {
        "name": "prompt_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in the prompt.",
        "children": [
          {
            "name": "cached_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Number of tokens served from cache."
          },
          {
            "name": "audio_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Audio input tokens present in the prompt."
          }
        ]
      },
      {
        "name": "seconds",
        "type": "integer",
        "isBold": true,
        "description": "Audio duration (seconds)."
      }
    ]
  }
]`} />

## Chat response chunk object (streaming output) 
<InlineSchemaV2 schema={`[
  {
    "name": "choices",
    "type": "array",
    "isBold": true,
    "description": "A list of chat completion choices.",
    "children": [
      {
        "name": "delta",
        "type": "object",
        "isBold": true,
        "description": "A chat completion delta generated by streamed model responses.",
        "children": [
          {
            "name": "content",
            "type": "string",
            "isBold": true,
            "description": "The contents of the chunk message."
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "description": "The role of the author of this message."
          }
        ]
      },
      {
        "name": "finish_reason",
        "type": [
          "string",
          "null"
        ],
        "isBold": true,
        "description": "The reason the model stopped generating tokens:<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">stop</code>: The model reached a natural stop point or a user‑provided stop sequence</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">length</code>: Terminated due to exceeding the model's maximum generation length</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">content_filter</code>: Content was omitted due to a content filter flag</li></ul>"
      },
      {
        "name": "index",
        "type": "integer",
        "isBold": true,
        "description": "The index of the choice in the list of choices."
      }
    ]
  },
  {
    "name": "created",
    "type": "integer",
    "isBold": true,
    "description": "The Unix timestamp (in seconds) of when the chat completion was created. Each chunk has the same timestamp."
  },
  {
    "name": "id",
    "type": "string",
    "isBold": true,
    "description": "A unique identifier for the chat completion.  Each chunk has the same ID."
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "description": "The model to generate the completion."
  },
  {
    "name": "object",
    "type": "string",
    "isBold": true,
    "description": "The object type, which is always <code class=\\"schema-inline-code\\">chat.completion.chunk</code>."
  },
  {
    "name": "usage",
    "type": [
      "object",
      "null"
    ],
    "isBold": true,
    "description": "Usage statistics for the completion request.",
    "children": [
      {
        "name": "completion_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the generated completion."
      },
      {
        "name": "prompt_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the prompt."
      },
      {
        "name": "total_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Total number of tokens used in the request (prompt + completion)."
      },
      {
        "name": "completion_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in a completion.",
        "children": [
          {
            "name": "reasoning_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Tokens generated by the model for reasoning. Always <code class=\\"schema-inline-code\\">0</code>."
          }
        ]
      },
      {
        "name": "prompt_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in the prompt.",
        "children": [
          {
            "name": "cached_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Number of tokens served from cache."
          },
          {
            "name": "audio_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Audio input tokens present in the prompt."
          }
        ]
      },
      {
        "name": "seconds",
        "type": "integer",
        "isBold": true,
        "description": "Audio duration (seconds)."
      }
    ]
  }
]`} />


--- DOCUMENT: Speech Synthesis (MiMo-TTS Series) - OpenAI API Compatibility ---
URL: https://mimo.mi.com/static/docs/api/audio/tts.md

# Speech Synthesis (MiMo-TTS Series) - OpenAI API Compatibility

## Request Address

```bash
https://api.xiaomimimo.com/v1/chat/completions
```

## Request Headers

{/* feishu-style:text-align:left */}
The API supports the following two authentication methods. Please choose one and add it to the request headers:

<Tab>
  <TabItem label={`API Key Authentication`}>

```json
api-key: $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
  <TabItem label={`Bearer Authentication`}>

```json
Authorization: Bearer $MIMO_API_KEY
Content-Type: application/json
```

  </TabItem>
</Tab>

## Request body 

<InlineSchemaV2 schema={`[
  {
    "name": "messages",
    "type": "array",
    "isBold": true,
    "required": true,
    "description": "The current conversation message list.",
    "children": [
      {
        "name": "User message",
        "type": "object",
        "isBold": false,
        "description": "Messages sent by an end user, containing prompts or additional context information.<br /><blockquote class=\\"schema-blockquote\\">Note: When generating audio using the <code class=\\"schema-inline-code\\">mimo-v2.5-tts-voicedesign</code> model, this message is required and is used to specify the text describing the voice design.</blockquote>",
        "children": [
          {
            "name": "content",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The contents of the user message."
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Role of the message author.<br />Available options: <code class=\\"schema-inline-code\\">user</code>"
          }
        ]
      },
      {
        "name": "Assistant message",
        "type": "object",
        "isBold": false,
        "description": "Messages sent by the model in response to user messages.<br /><blockquote class=\\"schema-blockquote\\">Note: When using the <code class=\\"schema-inline-code\\">mimo-v2.5-tts-voicedesign</code> model and <code class=\\"schema-inline-code\\">optimize_text_preview</code> is <code class=\\"schema-inline-code\\">true</code>, the assistant message is optional; in other cases, it is required.</blockquote>",
        "children": [
          {
            "name": "content",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "The contents of the assistant message, which is used to specify the target text for audio synthesis."
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "required": true,
            "description": "Role of the message author.<br />Available options: <code class=\\"schema-inline-code\\">assistant</code>"
          }
        ]
      }
    ]
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "required": true,
    "description": "Model ID is used to generate the response.<br />Available options: <code class=\\"schema-inline-code\\">mimo-v2.5-tts</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-tts-voicedesign</code>, <code class=\\"schema-inline-code\\">mimo-v2.5-tts-voiceclone</code>"
  },
  {
    "name": "audio",
    "type": "object",
    "isBold": true,
    "required": false,
    "description": "Parameters for audio output. For details, please refer to <a target=\\"_blank\\" rel=\\"noopener noreferrer\\" href=\\"https://mimo.mi.com/docs/en-US/quick-start/usage-guide/audio/speech-synthesis-v2.5\\">Speech Synthesis</a>.<br /><blockquote class=\\"schema-blockquote\\">Note: To generate audio, you must add a message with role set to <code class=\\"schema-inline-code\\">assistant</code>, which needs to specify the text for speech synthesis. Additionally, when using the <code class=\\"schema-inline-code\\">mimo-v2.5-tts-voicedesign</code> model, a message with the role of <code class=\\"schema-inline-code\\">user</code> is required. If <code class=\\"schema-inline-code\\">optimize_text_preview</code> is set to <code class=\\"schema-inline-code\\">true</code>, the <code class=\\"schema-inline-code\\">assistant</code> message can be omitted.</blockquote>",
    "children": [
      {
        "name": "format",
        "type": "string",
        "isBold": true,
        "required": false,
        "defaultValue": "wav",
        "description": "Specifies the output audio format. Default: <code class=\\"schema-inline-code\\">wav</code>, or <code class=\\"schema-inline-code\\">pcm</code> when you set <code class=\\"schema-inline-code\\">stream: true</code>.<br /><blockquote class=\\"schema-blockquote\\">Passing in <code class=\\"schema-inline-code\\">pcm</code> or <code class=\\"schema-inline-code\\">pcm16</code> both indicate specifying the use of the <code class=\\"schema-inline-code\\">pcm16</code> format.</blockquote>Available options: <code class=\\"schema-inline-code\\">wav</code>, <code class=\\"schema-inline-code\\">mp3</code>, <code class=\\"schema-inline-code\\">pcm</code>, <code class=\\"schema-inline-code\\">pcm16</code>"
      },
      {
        "name": "optimize_text_preview",
        "type": "boolean",
        "isBold": true,
        "required": false,
        "defaultValue": "false",
        "description": "Enables intelligent optimization of the target audio broadcast text.<br />When set to <code class=\\"schema-inline-code\\">true</code>, the input target text is intelligently polished; if no target text is provided, a broadcast-adapted target text is automatically generated. The finalized processed text is then fed into the model for speech synthesis.<br /><blockquote class=\\"schema-blockquote\\">Note: When this parameter is set to <code class=\\"schema-inline-code\\">true</code>, the <code class=\\"schema-inline-code\\">assistant</code> role message for specifying speech synthesis content can be omitted.</blockquote><blockquote class=\\"schema-blockquote\\">Currently, only the <code class=\\"schema-inline-code\\">mimo-v2.5-tts-voicedesign</code> model is supported.</blockquote>"
      },
      {
        "name": "voice",
        "type": "string",
        "isBold": true,
        "description": "The voice ID of the built-in voice or the base64 encoding of the audio sample.<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5-tts</code>: This field is optional and only supports using built-in voices, with the default value being <code class=\\"schema-inline-code\\">mimo_default</code></li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5-tts-voiceclone</code>: This field is required and only supports passing in the base64 encoding of audio samples, and only supports passing in audio sample files in <code class=\\"schema-inline-code\\">mp3</code> and <code class=\\"schema-inline-code\\">wav</code> formats</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5-tts-voicedesign</code> does not support this field</li></ul>Available options:<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">mimo-v2.5-tts</code>: <code class=\\"schema-inline-code\\">mimo_default</code>, <code class=\\"schema-inline-code\\">冰糖</code>, <code class=\\"schema-inline-code\\">茉莉</code>, <code class=\\"schema-inline-code\\">苏打</code>, <code class=\\"schema-inline-code\\">白桦</code>, <code class=\\"schema-inline-code\\">Mia</code>, <code class=\\"schema-inline-code\\">Chloe</code>, <code class=\\"schema-inline-code\\">Milo</code>, <code class=\\"schema-inline-code\\">Dean</code></li></ul>"
      }
    ]
  },
  {
    "name": "stream",
    "type": "boolean",
    "isBold": true,
    "required": false,
    "defaultValue": "false",
    "description": "If set to true, the model response data will be streamed to the client as it is generated using server-sent events."
  }
]`} />

## Chat response object (non-streaming output) 

<InlineSchemaV2 schema={`[
  {
    "name": "choices",
    "type": "array",
    "isBold": true,
    "description": "A list of chat completion choices.",
    "children": [
      {
        "name": "finish_reason",
        "type": "string",
        "isBold": true,
        "description": "The reason the model stopped generating tokens:<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">stop</code>: The model reached a natural stop point or a user‑provided stop sequence</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">length</code>: Terminated due to exceeding the model's maximum generation length</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">content_filter</code>: Content was omitted due to a content filter flag</li></ul>"
      },
      {
        "name": "index",
        "type": "integer",
        "isBold": true,
        "description": "The index of the choice in the list of choices."
      },
      {
        "name": "message",
        "type": "object",
        "isBold": true,
        "description": "A chat completion message generated by the model.",
        "children": [
          {
            "name": "content",
            "type": "string",
            "isBold": true,
            "description": "The contents of the message."
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "description": "The role of the author of this message."
          },
          {
            "name": "audio",
            "type": "object",
            "isBold": true,
            "description": "If the audio output is requested, this object contains data about the audio response from the model.",
            "children": [
              {
                "name": "id",
                "type": "string",
                "isBold": true,
                "description": "Unique identifier for this audio response."
              },
              {
                "name": "data",
                "type": "string",
                "isBold": true,
                "description": "Base64 encoded audio bytes generated by the model, in the format specified in the request."
              },
              {
                "name": "expires_at",
                "type": [
                  "number",
                  "null"
                ],
                "isBold": true,
                "description": "The Unix timestamp (in seconds) for when this audio response expires. Currently always <code class=\\"schema-inline-code\\">null</code>."
              },
              {
                "name": "transcript",
                "type": [
                  "string",
                  "null"
                ],
                "isBold": true,
                "description": "Transcript of the audio generated by the model. Currently always <code class=\\"schema-inline-code\\">null</code>."
              }
            ]
          },
          {
            "name": "final_text_preview",
            "type": "string",
            "isBold": true,
            "description": "The final audio broadcast text after intelligent optimization and polishing. This field is only returned when the request parameter <code class=\\"schema-inline-code\\">optimize_text_preview</code> is set to <code class=\\"schema-inline-code\\">true</code>."
          }
        ]
      }
    ]
  },
  {
    "name": "created",
    "type": "integer",
    "isBold": true,
    "description": "The Unix timestamp (in seconds) of when the chat completion was created."
  },
  {
    "name": "id",
    "type": "string",
    "isBold": true,
    "description": "A unique identifier for the chat completion."
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "description": "The model to generate the completion."
  },
  {
    "name": "object",
    "type": "string",
    "isBold": true,
    "description": "The object type, which is always <code class=\\"schema-inline-code\\">chat.completion</code>."
  },
  {
    "name": "usage",
    "type": [
      "object",
      "null"
    ],
    "isBold": true,
    "description": "Usage statistics for the completion request.",
    "children": [
      {
        "name": "completion_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the generated completion."
      },
      {
        "name": "prompt_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the prompt."
      },
      {
        "name": "total_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Total number of tokens used in the request (prompt + completion)."
      },
      {
        "name": "completion_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in a completion.",
        "children": [
          {
            "name": "reasoning_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Tokens generated by the model for reasoning. Always <code class=\\"schema-inline-code\\">0</code>."
          }
        ]
      },
      {
        "name": "prompt_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in the prompt.",
        "children": [
          {
            "name": "cached_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Number of tokens served from cache."
          }
        ]
      }
    ]
  }
]`} />

## Chat response chunk object (streaming output) 
<InlineSchemaV2 schema={`[
  {
    "name": "choices",
    "type": "array",
    "isBold": true,
    "description": "A list of chat completion choices.",
    "children": [
      {
        "name": "delta",
        "type": "object",
        "isBold": true,
        "description": "A chat completion delta generated by streamed model responses.",
        "children": [
          {
            "name": "content",
            "type": "string",
            "isBold": true,
            "description": "The contents of the chunk message."
          },
          {
            "name": "role",
            "type": "string",
            "isBold": true,
            "description": "The role of the author of this message."
          },
          {
            "name": "audio",
            "type": [
              "object",
              "null"
            ],
            "isBold": true,
            "description": "If the audio output modality is requested, this object contains data about the audio response from the model.",
            "children": [
              {
                "name": "id",
                "type": "string",
                "isBold": true,
                "description": "Unique identifier for this audio response."
              },
              {
                "name": "data",
                "type": "string",
                "isBold": true,
                "description": "Base64 encoded audio bytes generated by the model, in the format specified in the request."
              },
              {
                "name": "expires_at",
                "type": [
                  "number",
                  "null"
                ],
                "isBold": true,
                "description": "The Unix timestamp (in seconds) for when this audio response expires. Currently always <code class=\\"schema-inline-code\\">null</code>."
              },
              {
                "name": "transcript",
                "type": [
                  "string",
                  "null"
                ],
                "isBold": true,
                "description": "Transcript of the audio generated by the model. Currently always <code class=\\"schema-inline-code\\">null</code>."
              }
            ]
          },
          {
            "name": "final_text_preview",
            "type": "string",
            "isBold": true,
            "description": "The final audio broadcast text after intelligent optimization and polishing. This field is only returned when the request parameter <code class=\\"schema-inline-code\\">optimize_text_preview</code> is set to <code class=\\"schema-inline-code\\">true</code>."
          }
        ]
      },
      {
        "name": "finish_reason",
        "type": [
          "string",
          "null"
        ],
        "isBold": true,
        "description": "The reason the model stopped generating tokens:<br /><ul class=\\"schema-list\\"><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">stop</code>: The model reached a natural stop point or a user‑provided stop sequence</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">length</code>: Terminated due to exceeding the model's maximum generation length</li><li class=\\"schema-list-item\\"><code class=\\"schema-inline-code\\">content_filter</code>: Content was omitted due to a content filter flag</li></ul>"
      },
      {
        "name": "index",
        "type": "integer",
        "isBold": true,
        "description": "The index of the choice in the list of choices."
      }
    ]
  },
  {
    "name": "created",
    "type": "integer",
    "isBold": true,
    "description": "The Unix timestamp (in seconds) of when the chat completion was created. Each chunk has the same timestamp."
  },
  {
    "name": "id",
    "type": "string",
    "isBold": true,
    "description": "A unique identifier for the chat completion. Each chunk has the same ID."
  },
  {
    "name": "model",
    "type": "string",
    "isBold": true,
    "description": "The model to generate the completion."
  },
  {
    "name": "object",
    "type": "string",
    "isBold": true,
    "description": "The object type, which is always <code class=\\"schema-inline-code\\">chat.completion.chunk</code>."
  },
  {
    "name": "usage",
    "type": [
      "object",
      "null"
    ],
    "isBold": true,
    "description": "Usage statistics for the completion request.",
    "children": [
      {
        "name": "completion_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the generated completion."
      },
      {
        "name": "prompt_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Number of tokens in the prompt."
      },
      {
        "name": "total_tokens",
        "type": "integer",
        "isBold": true,
        "description": "Total number of tokens used in the request (prompt + completion)."
      },
      {
        "name": "completion_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in a completion.",
        "children": [
          {
            "name": "reasoning_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Tokens generated by the model for reasoning. Always <code class=\\"schema-inline-code\\">0</code>."
          }
        ]
      },
      {
        "name": "prompt_tokens_details",
        "type": "object",
        "isBold": true,
        "description": "Breakdown of tokens used in the prompt.",
        "children": [
          {
            "name": "cached_tokens",
            "type": "integer",
            "isBold": true,
            "description": "Number of tokens served from cache."
          }
        ]
      }
    ]
  }
]`} />


--- DOCUMENT: List Models ---
URL: https://mimo.mi.com/static/docs/api/model/list-models.md

# List Models

## Request Address

```bash
https://api.xiaomimimo.com/v1/models
```

## Request Headers

{/* feishu-style:text-align:left */}
The API supports the following two authentication methods. Please choose one and add it to the request headers:

<Tab>
  <TabItem label={`API Key Authentication`}>

```json
api-key: $MIMO_API_KEY
```

  </TabItem>
  <TabItem label={`Bearer Authentication`}>

```json
Authorization: Bearer $MIMO_API_KEY
```

  </TabItem>
</Tab>

## Response
<InlineSchemaV2 schema={`[
  {
    "name": "object",
    "type": "string",
    "isBold": true,
    "description": "The object type.<br />Available options: <code class=\\"schema-inline-code\\">list</code>"
  },
  {
    "name": "data",
    "type": "array",
    "isBold": true,
    "description": "An array of model objects.",
    "children": [
      {
        "name": "id",
        "type": "string",
        "isBold": true,
        "description": "The model identifier, which can be referenced in the API endpoints."
      },
      {
        "name": "object",
        "type": "string",
        "isBold": true,
        "description": "The object type.<br />Available options: <code class=\\"schema-inline-code\\">model</code>"
      },
      {
        "name": "owned_by",
        "type": "string",
        "isBold": true,
        "description": "The organization that owns the model."
      }
    ]
  }
]`} />


--- DOCUMENT: Pay‑As‑You‑Go API ---
URL: https://mimo.mi.com/static/docs/price/pay-as-you-go.md

# API Pricing

{/* feishu-style:text-align:left */}
**Pay-as-you-go for the Xiaomi MiMo API uses the ordinary API Key of the Open Platform and consumes the account balance based on the actual Token usage, which is not interoperable with the Token Plan package quota.** 

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**Billing Instructions** 

- Billing Unit: China：RMB / M tokens;  Overseas: dollar / M tokens

- Cache Hit: When the requested prefix content hits the Prompt Cache, it will be billed according to the cached hit price.

- Cache Write: Limited-time Free

- ASR series models are billed based on the duration of the input audio: duration statistics are accurate to the second, and ultimately converted to hourly billing

- Internet search is billed independently based on the number of calls and is not included in the Token price

</div>

### Domestic Pricing of the Model

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

- `mimo-v2.5-pro` and `mimo-v2.5` **will be officially deprecated at 10:00 (Beijing time) on October 21,2026. It is recommended to switch to the new version of the models as soon as possible. For details, please refer to** [Model Deprecation](https://mimo.mi.com/docs/zh-CN/updates/deprecate).

</div>

{/* feishu-style:text-align:left */}
**Language Models**

<table>
<colgroup>
<col style="width: 13.46%" />
<col style="width: 44.27%" />
<col style="width: 15.91%" />
<col style="width: 16.91%" />
<col style="width: 9.450000000000003%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Inference Type**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Model Name**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Input (Cache Hit)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Input (Cache Miss)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Output**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td rowspan="3"><span style="display: inline-block; width: 100%; text-align: left;">**Real-time API**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`、`mimo-v2.5-pro`(to be deprecated)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.025</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥3.00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥6.00</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`、`mimo-v2.5`(to be deprecated)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.02</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥1.00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥2.00</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro-ultraspeed`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.25</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥30.00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥60.00</span></td>
</tr>
<tr>
<td rowspan="2"><span style="display: inline-block; width: 100%; text-align: left;">**Batch API**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.0125</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥1.50</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥3.00</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.01</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.50</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥1.00</span></td>
</tr>
</tbody>
</table>

> `mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro`, `mimo-v2.5 `do not support batch API.

{/* feishu-style:text-align:left */}
**ASR Series**

<table>
<colgroup>
<col style="width: 41.64%" />
<col style="width: 58.36%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Model Name**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Input audio duration**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-asr`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥0.5 /h</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
**TTS Series**

{/* feishu-style:text-align:left */}
`mimo-v2.5-tts`, `mimo-v2.5-tts-voiceclone`, `mimo-v2.5-tts-voicedesign` are free for a limited time

{/* feishu-style:text-align:left */}
<br></br>

### Overseas Pricing of the Model 

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

- `mimo-v2.5-pro` and `mimo-v2.5` **will be officially deprecated at 10:00 (Beijing time) on October 21,2026. It is recommended to switch to the new version of the models as soon as possible. For details, please refer to** [Model Deprecation](https://mimo.mi.com/docs/zh-CN/updates/deprecate).

</div>

{/* feishu-style:text-align:left */}
**Language Models**

<table>
<colgroup>
<col style="width: 13.7%" />
<col style="width: 44.1%" />
<col style="width: 15.92%" />
<col style="width: 16.93%" />
<col style="width: 9.349999999999994%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Inference Type**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Model Name**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Input (Cache Hit)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Input (Cache Miss)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Output**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td rowspan="3"><span style="display: inline-block; width: 100%; text-align: left;">**Real-time API**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`、`mimo-v2.5-pro`(to be deprecated)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.0036</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.435</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.87</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`、`mimo-v2.5`(to be deprecated)</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.0028</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.14</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.28</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro-ultraspeed`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.036</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$4.35</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$8.7</span></td>
</tr>
<tr>
<td rowspan="2"><span style="display: inline-block; width: 100%; text-align: left;">**Batch API**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-pro`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.0018</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.2175</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.435</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.6-flash`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.0014</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.07</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.14</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
`mimo-v2.6-pro-ultraspeed`, `mimo-v2.5-pro`, `mimo-v2.5 `do not support batch API.

{/* feishu-style:text-align:left */}
**ASR Series**

<table>
<colgroup>
<col style="width: 41.64%" />
<col style="width: 58.36%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Model Name**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Input audio duration**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">`mimo-v2.5-asr`</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$0.074 /h</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
**TTS Series**

{/* feishu-style:text-align:left */}
`mimo-v2.5-tts`, `mimo-v2.5-tts-voiceclone`, `mimo-v2.5-tts-voicedesign` are free for a limited time

{/* feishu-style:text-align:left */}
<br></br>

### Pricing for Web Search Plugins 

<table>
<colgroup>
<col style="width: 21.97%" />
<col style="width: 21.97%" />
<col style="width: 56.06%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Service Item**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Price**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Description**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Domestic Internet Connectivity Service</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">¥16 /1000 times</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Includes web search and web parsing, used for searching relevant content in domestic regional network connections </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Overseas Internet Connectivity Service </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$5 /1000 times</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Includes web search and web parsing, used for networked search of relevant content in overseas regions </span></td>
</tr>
</tbody>
</table>


--- DOCUMENT: Token Plan ---
URL: https://mimo.mi.com/static/docs/price/token-plan.md

# Token Plan

{/* feishu-style:text-align:left */}
**Token Plan** is an exclusive subscription solution launched for AI programming scenarios. You can use the cost-effective subscription resource package to access MiMo's flagship large language model in various mainstream AI development tools. The platform currently offers two versions: the individual version and the team version.

## Core Strengths

- **Covers flagship models** — all plans support the latest flagship models mimo-v2.6-pro, mimo-v2.6-flash, as well as ASR and TTS models. Adopts a token conversion mechanism with transparent and controllable quota

- **Flexible Subscription Plans** — Tiered packages are available for both individual and team versions to meet the diverse development needs of individual and team developers

- **Multi-ecosystem Out Of The Box** — Compatible with mainstream development toolchains including OpenCode, OpenClaw, Claude Code, etc.

- **Exclusive Benefits for Team Plan** — In addition to inheriting all the benefits of the Individual Plan, users of the Team Plan can enjoy exclusive privileges including unified management of storage seats, team usage control and analysis, as well as centralized billing and invoice management

## Pricing and Quotas for Individual Edition

#### Monthly Plan

<table>
<colgroup>
<col style="width: 18.14%" />
<col style="width: 21.02%" />
<col style="width: 19.49%" />
<col style="width: 20.68%" />
<col style="width: 20.670000000000016%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Lite**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Pricing**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$6/month, ¥39/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$16/month, ¥99/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$50/month, ¥329/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$100/month, ¥659/month</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Fixed Monthly Quota**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">4.1 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">11 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">38 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">82 billion Credits</span></td>
</tr>
</tbody>
</table>

#### Annual Package

<table>
<colgroup>
<col style="width: 18.14%" />
<col style="width: 21.02%" />
<col style="width: 19.49%" />
<col style="width: 20.68%" />
<col style="width: 20.670000000000016%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Lite**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Pricing**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$63.36/year, ¥411.84/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 168.96/year, CNY 1045.44/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$528.00/year, ¥3474.24/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 1,056.00/year, CNY 6,959.04/year</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Annual Fixed Quota**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">49.2 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">132 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">456 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">984 billion Credits</span></td>
</tr>
</tbody>
</table>

## Pricing and Quotas for Team Edition

#### Monthly Plan

<table>
<colgroup>
<col style="width: 22.34%" />
<col style="width: 24.01%" />
<col style="width: 25.47%" />
<col style="width: 28.180000000000007%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Pricing**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$16/seat/month, ¥99/seat/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$50/seat/month, ¥329/seat/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$100/seat/month, ¥659/seat/month</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Fixed Monthly Quota**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">11 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">38 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">82 billion Credits</span></td>
</tr>
</tbody>
</table>

#### Annual Package

<table>
<colgroup>
<col style="width: 22.96%" />
<col style="width: 24.68%" />
<col style="width: 26.18%" />
<col style="width: 26.180000000000007%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Pricing**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 168.96/seat/year, CNY 1,044/seat/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 528/seat/year, CNY 3,468/seat/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 1,056/seat/year, CNY 6,948/seat/year</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Fixed monthly quota per seat**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">11 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">38 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">82 billion Credits</span></td>
</tr>
</tbody>
</table>

## Applicable Scenarios

<table>
<colgroup>
<col style="width: 17.94%" />
<col style="width: 20.79%" />
<col style="width: 19.28%" />
<col style="width: 20.45%" />
<col style="width: 21.539999999999992%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Lite (Individual Only)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Applicable Scenarios**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Ideal for first-time users who want to try lobster</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, it can execute approximately **200 rounds of medium-to-complex tasks**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Suitable for office workers who frequently use AI to boost their work efficiency</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, it can execute approximately **1600 rounds of medium-to-complex tasks**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Ideal for developers and professional productivity enthusiasts who use AI frequently on a daily basis</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, it can execute approximately **5600 rounds of medium-to-complex tasks**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Ideal for high-intensity, hardcore users who treat AI as a core productivity tool</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, it can execute approximately **12800 rounds of medium-to-complex tasks**</span></td>
</tr>
</tbody>
</table>

> The above describes the applicable scenario scope of the monthly plan, and the order of magnitude of task processing for the annual plan is approximately 12 times that of the monthly plan.

## Quota Consumption Rules

> The quota consumption rules for TokenPlan are the same for both the individual version and the team version

{/* feishu-style:text-align:left */}
Language models deduct Credits quota based on the number of tokens; ASR models deduct Credits quota according to the duration of the input audio (the duration is counted to the exact second and converted to hourly billing); TTS series models are free for a limited time and do not consume package Credits. The specific conversion rules are as follows.

{/* feishu-style:text-align:left */}
Language Model

<table>
<colgroup>
<col style="width: 19.37%" />
<col style="width: 28.49%" />
<col style="width: 31.2%" />
<col style="width: 20.939999999999998%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input (Cache Hit) Token</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input (cache miss) Token</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Output Token</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.6-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2.5 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">300 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">600 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.6-flash</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">200 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2.5 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">300 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">600 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">200 Credits</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
ASR Model

<table>
<colgroup>
<col style="width: 40.48%" />
<col style="width: 59.52%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input audio duration (h)</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-asr</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">30M Credits</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
The TTS series models are available for free for a limited time and will not consume your plan credits.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**The Individual edition additionally supports mimo-v2.5-pro and mimo-v2.5.Both of these two models will be officially taken offline at 10:00 on October 21, 2026 Beijing Time, and it is recommended to switch to the new version of the models as soon as possible.**

</div>

{/* feishu-style:text-align:left */}
For example, if you subscribe to the Lite plan (4.1B Credits), you can call the MiMo- V2.6 series models either individually or in combination. After you use 10M mimo-v2.6-pro input tokens (cache miss), this is equivalent to consuming 3000 M Credit s, and you will still have 1100M mimo-v2.5-flash Credits remaining.In addition, if you dedicate the entire quota of the Lite plan exclusively to the ASR model, you will get 4100M per month ÷ 30M/hour = 136.6 hours (equivalent to processing 4.5 hours of audio per day for 1 consecutive month). You can check the quota and usage of your current plan at [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage).

- **Quota Exhaustion:** When the monthly total quota of the package is used up, the system will suspend the service and will not continue to deduct from your bonus or account balance.

- **To continue using the service:**  **For the Individual Plan,**  please purchase an upgrade package to unlock new package resources, or switch to the standard API which charges based on per-token unit price, so that you can continue using the service without usage restrictions. **For the Team Plan**, please purchase additional seats and reallocate them, or switch to the standard API.

## Discount Offers

- First Purchase Discount: Users purchasing the Individual Edition for the first time can enjoy a 12% discount, and each account is only eligible for this discount once; the Team Edition is not eligible for this discount.

- Annual Subscription (Auto-renewal): Compared with the monthly auto-renewal plan, both the Individual Plan and Team Plan enjoy a 12% discount for annual auto-renewal, and first-purchase offers are not applicable to annual auto-renewal plans;

- Night Discount Rate: During off-peak hours (Beijing Time 00:00–08:00, i. e. UTC 16:00–24:00), the consumption coefficient is 0.8x.

## For more details

{/* feishu-style:text-align:left */}
For more information about TokenPlan, please refer to [TokenPlan](https://platform.xiaomimimo.com/#/docs/faq).


--- DOCUMENT: Individual Subscription ---
URL: https://mimo.mi.com/static/docs/tokenplan/Token Plan/subscription.md

# Individual Subscription

{/* feishu-style:text-align:left */}
Token Plan Individual Edition is an exclusive subscription solution launched for individual developers in AI programming scenarios, allowing you to use the cost-effective subscription resource package to call MiMo's flagship large language model in various mainstream AI development tools.

## Core Strengths

- **Covers flagship models** — all plans support mimo-v2.6-pro, mimo-v2.6-flash, mimo-v2.5-pro, mimo-v2.5, mimo-v2.5-asr, mimo-v2.5-tts-voiceclone, mimo-v2.5-tts-voicedesign, mimo-v2.5-tts, a total of 8 models. It adopts a token conversion mechanism with transparent and controllable quotas.

- **Flexible Subscription Plan** — Four-tier packages to meet various development needs of individual developers

- **Multi-ecosystem Out Of The Box** — Compatible with mainstream development toolchains including OpenCode, OpenClaw, Claude Code, etc.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**The two models mimo-v2.5-pro and mimo-v2.5 will be officially taken offline at 10:00 (Beijing time) on October 21, 2026. It is recommended to switch to the new version of the models as soon as possible.**

</div>

## Quota Limit

#### Monthly Plan

<table>
<colgroup>
<col style="width: 18.14%" />
<col style="width: 21.02%" />
<col style="width: 19.49%" />
<col style="width: 20.68%" />
<col style="width: 20.670000000000016%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Lite**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Pricing**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$6/month, ¥39/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$16/month, ¥99/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$50/month, ¥329/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$100/month, ¥659/month</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Fixed Monthly Quota**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">4.1 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">11 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">38 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">82 billion Credits</span></td>
</tr>
</tbody>
</table>

#### Annual Package

<table>
<colgroup>
<col style="width: 18.14%" />
<col style="width: 21.02%" />
<col style="width: 19.49%" />
<col style="width: 20.68%" />
<col style="width: 20.670000000000016%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Lite**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Pricing**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$63.36/year, ¥411.84/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$168.96/year, ¥1045.44/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 528.00/year, CNY 3474.24/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 1,056.00/year, CNY 6,959.04/year</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Annual Fixed Quota**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">49.2 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">132 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">456 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">984 billion Credits</span></td>
</tr>
</tbody>
</table>

#### Applicable Scenarios

<table>
<colgroup>
<col style="width: 17.94%" />
<col style="width: 20.79%" />
<col style="width: 19.28%" />
<col style="width: 20.45%" />
<col style="width: 21.539999999999992%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Lite**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Applicable Scenarios**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Ideal for first-time users who want to try lobster</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, it can execute approximately **200 rounds of medium-to-complex tasks**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Ideal for office professionals who regularly use AI to boost their work efficiency</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, it can execute approximately **1600 rounds of medium-to-complex tasks**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Ideal for developers and professional productivity enthusiasts who use AI frequently on a daily basis</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, it can execute approximately **5600 rounds of medium-to-complex tasks**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Ideal for high-intensity, hardcore users who treat AI as a core productivity tool</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, it can execute approximately **12800 rounds of medium-to-complex tasks**</span></td>
</tr>
</tbody>
</table>

> The above describes the applicable scenario scope of the monthly package; the task processing magnitude of the annual package is approximately 12 times that of the monthly package.

#### Quota Consumption Rules

{/* feishu-style:text-align:left */}
Language models deduct Credit quota based on the number of tokens; ASR models deduct Credit quota according to the duration of the input audio (the duration is counted to the nearest second and finally converted to hours for statistics); TTS series models are free for a limited time and do not consume package Credit. The specific conversion rules are as follows.

{/* feishu-style:text-align:left */}
Language Model

<table>
<colgroup>
<col style="width: 19.37%" />
<col style="width: 28.49%" />
<col style="width: 31.2%" />
<col style="width: 20.939999999999998%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input (Cache Hit) Token</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input (cache miss) Token</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Output Token</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.6-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2.5 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">300 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">600 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.6-flash</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">200 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2.5 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">300 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">600 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">200 Credits</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
ASR Model

<table>
<colgroup>
<col style="width: 40.48%" />
<col style="width: 59.52%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input audio duration (h)</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-asr</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">30M Credits</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
The TTS series models are available for free for a limited time and will not consume your package credits.

{/* feishu-style:text-align:left */}
For example, if you subscribe to the Lite plan (4.1B Credits), you can call MiMo- V2.6 series models either individually or in combination. After you use 10M input (cache miss) tokens of mimo-v2.6-pro, this is equivalent to consuming 3000M Credit s, and you will still have a remaining Credits quota of 1100M for mimo-v2.6-flash.In addition, if you dedicate the entire quota of the Lite plan exclusively to the ASR model, you will get 4100M ÷ 30M/hour = 136.6 hours per month (equivalent to processing 4.5 hours of audio per day for consecutive use over 1 month). You can check the quota and usage of your current plan at [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage).

- **Quota Exhaustion:** When the monthly total quota of your plan is used up, the system will suspend the service and will not continue to deduct from your bonus or account balance.

- **To continue using the service:** please purchase an upgrade plan to unlock new plan resources, or switch to the standard API which is billed based on per-token unit price, so that you can continue using the service without usage restrictions.

## Package Purchase

- **Discounts: 12% off for your first package purchase, 12% off for annual auto-renewal, and 20% lower consumption during nighttime**

   - First Purchase Discount: Enjoy a 12% discount on your first purchase, with only one discount eligible per account;

   - Annual auto-renewal plan: Enjoy a 12% discount compared to the monthly auto-renewal plan; first-purchase offers are not applicable to the annual plan.

   - Off-peak discount rate: During non-peak hours (00:00–08:00 Beijing Time, equivalent to 16:00–24:00 UTC), a consumption coefficient of 0.8x applies.

- **Top-up upgrade for plans by paying price difference: Currently, the platform only supports purchasing one individual plan at a time. If you wish to get more quota before your current plan expires**, the Credit quota you have already used will be converted into an equivalent amount, and you can pay the price difference on this basis to upgrade to a higher plan for more Credits. Cross-tier plan upgrade via price difference top-up is supported, while plan downgrade is not allowed. If you have already upgraded to the top-tier Max plan, further upgrade is unavailable. **After your plan expires, you can repurchase a plan of any tier.** 

> Price difference = New package price -(Remaining balance of original package / Total amount of original package) * Original package price

- **Auto-renewal supported**: The auto-renewal feature has been launched. You can enjoy a discount when you first sign up for a continuous subscription. Stay tuned for more updates.

- **Refunds are not supported for the time being**: Please note that once a subscription service is purchased, it takes effect immediately and no refunds will be provided. Unused quotas within the package are also non-refundable. Please carefully select an appropriate subscription plan based on your own usage needs.

- **Invoice support**: Domestic users can apply for invoices based on the transaction orders in their recharge details, and the actual invoiceable amount is the actual amount paid. Overseas users can directly download the invoice after purchase, or download it on the recharge details page.

## Package Usage

{/* feishu-style:text-align:left */}
The quota of the Token Plan package is only available for use in programming tools (such as OpenClaw, OpenCode, etc.), and it is prohibited to use it in the form of API calls for request behaviors in obvious non-coding scenarios such as automated scripts and custom application backends.

{/* feishu-style:text-align:left */}
If calls beyond the permitted scope are made using the API Key corresponding to the package, such act will be deemed as a violation or abuse, and the platform reserves the right to take measures including but not limited to suspending services for the relevant subscription and blocking the API Key.

## Quick Guide

{/* feishu-style:text-align:left */}
Get started quickly with Token Plan, from subscribing to a package to using the MiMo model in coding tools.

### Subscribe to the Token Plan

{/* feishu-style:text-align:left */}
Visit [Token Plan](https://platform.xiaomimimo.com/#/token-plan), then select and purchase a suitable subscription package as needed.

### Obtain the package-specific Base URL and API Key

{/* feishu-style:text-align:left */}
After successful subscription, you can go to the [Token Plan ](https://platform.xiaomimimo.com/#/console/plan-manage)page to obtain the package-specific Base URL and API Key.

- **API Key**: On the [Token Plan ](https://platform.xiaomimimo.com/#/console/plan-manage)page, obtain your exclusive API Key (in the format of `tp-xxxxx`).

- **Base URL**: You will need to configure one of the following Base URLs in your AI programming tool later (**The protocol varies by tool, and the Base URL shall be subject to what is displayed on the** [**Token Plan**](https://platform.xiaomimimo.com/#/console/plan-manage) **page**). For specific operations, please refer to the user guide document of the corresponding AI programming tool.

   - **OpenAI Compatible Protocol**

      - China Cluster: `https://token-plan-cn.xiaomimimo.com/v1`

      - Singapore Cluster: `https://token-plan-sgp.xiaomimimo.com/v1`

      - European Cluster: `https://token-plan-ams.xiaomimimo.com/v1`

   - **Anthropic Compatibility Protocol**

      - China Cluster: `https://token-plan-cn.xiaomimimo.com/anthropic` 

      - Singapore Cluster: `https://token-plan-sgp.xiaomimimo.com/anthropic`

      - European Cluster: `https://token-plan-ams.xiaomimimo.com/anthropic`

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**Precautions**

- Please keep your API Key properly and do not disclose it to others.

- The API Key is only valid during the validity period of the Token Plan package you have subscribed to.

</div>

## Application in AI Agents and Programming Tools

{/* feishu-style:text-align:left */}
The Token Plan is supported for use in multiple mainstream AI programming tools, and all tools share the usage quota of the subscribed package.

{/* feishu-style:text-align:left */}
Go to [AI Tools Overview ](https://platform.xiaomimimo.com/#/docs/integration/tools-overview)to view the configuration guide corresponding to the tools you are using (such as OpenCode, OpenClaw, etc.).

## Frequently Asked Questions

{/* feishu-style:text-align:left */}
For more frequently asked questions, please refer to [FAQ](https://mimo.mi.com/docs/zh-CN/quick-start/faq/token-plan/Plans&Pricing).

## Individual Subscription

{/* feishu-style:text-align:left */}
Token Plan 个人版是面向个人开发者的 AI 编程场景推出的专属订阅方案，您可使用高性价比的订阅资源包，在各类主流 AI 开发工具中调用 MiMo 旗舰大模型。


--- DOCUMENT: Team Subscription ---
URL: https://mimo.mi.com/static/docs/tokenplan/Token Plan/team.md

# Team Subscription

{/* feishu-style:text-align:left */}
**Token Plan Team Edition** is an AI programming subscription solution tailored for development teams and enterprises. Building on all the benefits of the individual edition, it provides capabilities including member management, seat allocation, team usage statistics, and centralized billing management. The team owner purchases packages and allocates seats in a unified manner, and members can call the MiMo model in mainstream AI programming and agent tools via their exclusive API Keys.

## Core Strengths

- **First to integrate the latest flagship models and features:** All tiered plans support a total of 6 models, namely mimo-v2.6-pro, mimo-v2.6-flash, mimo-v2.5-asr, mimo-v2.5-tts-voiceclone, mimo-v2.5-tts-voicedesign and mimo-v2.5-tts.

- **Unified Space Seat Management:** Centralize the management of team members, and allocate, reclaim and purchase additional seats as needed to flexibly adapt to team changes.

- **Team usage is transparent and controllable:** Independent metering based on seats, with support for viewing usage by model and member, so as to clearly grasp the consumption of team quota.

- **Centralized Billing and Invoice Management:** View records of plan purchase, renewal and seat add-on in a unified manner, apply for invoices centrally to streamline financial management.

- **Flexible Subscription Plans:** Three tiered packages including Standard, Pro and Max, with support for monthly auto-renewal, annual auto-renewal and one-time purchase, to meet the needs of teams of different sizes and usage intensities.

- **All benefits of the Personal Edition are inherited:** Multi-ecosystem Out Of The Box, enjoy off-peak discounts during nighttime, with no 5-hour cap or weekly usage limit.

## Quota Limit

#### Monthly Plan

<table>
<colgroup>
<col style="width: 15.5%" />
<col style="width: 26.37%" />
<col style="width: 29.07%" />
<col style="width: 29.060000000000002%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Pricing**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">$16/seat/month, ¥99/seat/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 50/seat/month, CNY 329/seat/month</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 100/seat/month, CNY 659/seat/month</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Fixed monthly quota per seat**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">11 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">38 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">82 billion Credits</span></td>
</tr>
</tbody>
</table>

#### Annual Package

<table>
<colgroup>
<col style="width: 12.8%" />
<col style="width: 29.07%" />
<col style="width: 29.07%" />
<col style="width: 29.060000000000002%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Pricing**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 168.96/seat/year, CNY 1,044/seat/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 528/seat/year, CNY 3,468/seat/year</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">USD 1,056/seat/year, CNY 6,948/seat/year</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Fixed monthly quota per seat**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">11 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">38 billion Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">82 billion Credits</span></td>
</tr>
</tbody>
</table>

> All subscription tiers **require a minimum purchase of 2 seats**. Annual plans are paid on a yearly basis, with seat quotas reset monthly rather than being allocated in full at the start of the year.

#### Applicable Scenarios

<table>
<colgroup>
<col style="width: 12.78%" />
<col style="width: 29.01%" />
<col style="width: 29.11%" />
<col style="width: 29.099999999999994%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Standard**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Pro**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Max**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Applicable Scenarios**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Team members suitable for using AI to improve efficiency in daily work</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, each seat can execute approximately **1600 rounds of medium-to-complex tasks**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Suitable for developers and professionals who use AI frequently on a daily basis</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, each seat can execute approximately **5600 rounds of medium-to-complex tasks**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Hardcore users who are suitable to use AI as a core productivity tool</span><br /><span style="display: inline-block; width: 100%; text-align: left;">Using mimo-v2.6-flash as the baseline, each seat can execute approximately **12,800 rounds of medium-to-complex tasks**</span></td>
</tr>
</tbody>
</table>

> The above is the scope of scenarios corresponding to the monthly quota per seat. The annual cumulative task processing volume of the annual package is approximately 12 times that of the monthly quota, and the quota is still reset on a monthly basis.

#### Quota Consumption Rules

{/* feishu-style:text-align:left */}
Language models deduct Credits quota based on the number of tokens; ASR models deduct Credits quota according to the duration of the input audio (the duration is counted to the exact second and converted to hourly billing); TTS series models are free for a limited time and do not consume package Credits. The specific conversion rules are as follows.

{/* feishu-style:text-align:left */}
**Language Model**

<table>
<colgroup>
<col style="width: 24.93%" />
<col style="width: 26.29%" />
<col style="width: 28.05%" />
<col style="width: 20.730000000000004%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input (Cache Hit) Token</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input (cache miss) Token</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Output Token</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.6-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2.5 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">300 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">600 Credits</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.6-flash</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">2 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100 Credits</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">200 Credits</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
**ASR Model**

<table>
<colgroup>
<col style="width: 48.02%" />
<col style="width: 51.98%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Input audio duration (h)</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-asr</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">30M Credits</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
**TTS model**

{/* feishu-style:text-align:left */}
mimo-v2.5-tts-voiceclone, mimo-v2.5-tts-voicedesign and mimo-v2.5-tts are available for free for a limited time, and will not consume the credits in your package.

> The Team version is metered independently per seat. When an allocated and consumed seat is reassigned, the new member inherits the remaining quota of that seat for the current billing cycle, and the already consumed portion will not be reset.

- **Quota Exhaustion:** When the monthly quota of a seat is used up, that seat will be suspended, without affecting other seats that still have remaining quota. **A team-shared usage package is coming soon, stay tuned.** 

- **To continue using the service:** you can contact the team owner to purchase additional seats as needed and reassign them, or switch to the regular pay-as-you-go API service to keep using it.

## Package Purchase

- **Discounts: Enjoy an approximate 12% discount for annual auto-renewal plans or one-time purchases valid for 1 year or longer, plus 20% lower resource consumption during nighttime hours**

   - Annual Discount: Annual recurring subscription enjoys an approximately 12% discount off the monthly rate, and one-time purchases covering 12 months are also eligible for the annual package price. No first-purchase discount is available for the Team Plan.

   - Off-peak discount rate: A consumption coefficient of 0.8x applies during off-peak hours (00:00–08:00 Beijing Time, 16:00–24:00 UTC).

- **Flexible Purchase Options:** We support monthly auto-renewal, annual auto-renewal, and one-time purchases, with a minimum order of 2 seats. For one-time purchases, you can choose a validity period of 1 to 12 months to meet the usage needs of teams with different cycle requirements.

- **Support for adding seats mid-subscription:** The owner may add seats of the same tier during the validity period of the current plan, with a minimum of 1 seat per addition. The fee and current quota for the newly added seats will be prorated based on the remaining validity period, and their expiration date will align with that of the current plan.

- **Auto-renewal supported:** Monthly and annual subscription plans are auto-renewable, which can be turned off by the subscriber at any time.

- **Invoice Support:** Domestic users can apply for invoices based on the transaction orders in their recharge details, and the actual invoiceable amount is the actual payment amount. Overseas users can directly download the invoice after purchase, or download it on the recharge details page.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**Precautions**

- Multiple members are not supported to share the same seat

- Upgrades and mixed purchase of package tiers are temporarily not supported during the validity period. If you need additional usage resources, you can purchase extra seats for the current tier as needed.

- Unsubscription and refund for individual or partial seats are not supported. No refund will be provided for unused quotas within the package. Please carefully select an appropriate subscription plan and number of seats based on your own usage needs.

</div>

## Package Usage

{/* feishu-style:text-align:left */}
The quota of the Token Plan package is only available for use in programming tools (such as OpenClaw, OpenCode, etc.), and it is prohibited to use it in the form of API calls for request behaviors in obvious non-coding scenarios such as automated scripts and custom application backends.

{/* feishu-style:text-align:left */}
If calls beyond the permitted scope are made using the API Key corresponding to the package, such act will be deemed as a violation or abuse, and the platform reserves the right to take measures including but not limited to suspending services for the relevant subscription and blocking the API Key.

## Quick Guide

{/* feishu-style:text-align:left */}
Get started quickly with Token Plan, from subscribing to a package to using the MiMo model in coding tools.

### Subscribe to the Token Plan

{/* feishu-style:text-align:left */}
Visit [Token Plan](https://platform.xiaomimimo.com/#/token-plan), and select and purchase the appropriate team edition subscription package as needed.

### Invite members and assign seats

{/* feishu-style:text-align:left */}
After successful subscription, you can invite members in the Team Edition TokenPlan Console, and assign available seats before you can use the package. Of course, you can also assign a seat to yourself.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AYVfb2vfNoZBTxxCrYScqdwVnIg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AYVfb2vfNoZBTxxCrYScqdwVnIg.png" width="2940" height="1480" alt="图片" style="margin: 16px auto;" />

### Obtain the package-specific Base URL and API Key

{/* feishu-style:text-align:left */}
After members successfully access the team workspace and obtain the seats assigned by the owner, they can acquire the package-specific Base URL and API Key within the team workspace (the owner should enter "My Seats").

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/U1Jbbn5NgojKwFx3lVHcNKZ7n4b.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/U1Jbbn5NgojKwFx3lVHcNKZ7n4b.png" width="2940" height="1398" alt="图片" style="margin: 16px auto;" />

- **API Key**: Obtain a dedicated API Key (in the format of `ttp-xxxxx`) within the team workspace.

- **Base URL**: You will need to configure one of the following Base URLs in the AI programming tool later (**The protocol varies by tool, and the Base URL shall be subject to what is displayed on the Team Edition TokenPlan page**). For specific operations, please refer to the corresponding user guide document for the AI programming tool.

   - **OpenAI Compatible Protocol**

      - China Cluster: `https://token-plan-cn.xiaomimimo.com/v1`

      - Singapore Cluster: `https://token-plan-sgp.xiaomimimo.com/v1`

      - European Cluster: `https://token-plan-ams.xiaomimimo.com/v1`

   - **Anthropic Compatibility Protocol**

      - China Cluster: `https://token-plan-cn.xiaomimimo.com/anthropic` 

      - Singapore Cluster: `https://token-plan-sgp.xiaomimimo.com/anthropic`

      - European Cluster: `https://token-plan-ams.xiaomimimo.com/anthropic`

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**Precautions**

- Please keep your API Key properly and do not disclose it to others.

- The API Key is only valid during the validity period of the Token Plan package you have subscribed to.

</div>

## Application in AI Agents and Programming Tools

{/* feishu-style:text-align:left */}
The Token Plan is supported for use in multiple mainstream AI programming tools, and all tools share the usage quota of the subscribed package.

{/* feishu-style:text-align:left */}
Go to [AI Tools Overview ](https://platform.xiaomimimo.com/#/docs/integration/tools-overview)to view the configuration guide corresponding to the tools you are using (such as OpenCode, OpenClaw, etc.).

## Frequently Asked Questions

{/* feishu-style:text-align:left */}
For more frequently asked questions, please refer to [FAQ](https://mimo.mi.com/docs/zh-CN/quick-start/faq/token-plan/Team-Benefits).


--- DOCUMENT: Quick Access ---
URL: https://mimo.mi.com/static/docs/tokenplan/Token Plan/quick-access.md

# Quick Access

{/* feishu-style:text-align:left */}
This article describes how to quickly integrate with Token Plan. The entire process from subscription to invocation can be completed in just 3 steps.

## Step 1: Subscribe to the Token Plan

{/* feishu-style:text-align:left */}
Go to [Token Plan ](https://platform.xiaomimimo.com/#/token-plan)and select a suitable subscription plan.

## Step 2: Obtain the voucher

{/* feishu-style:text-align:left */}
After successful subscription, please go to [Token Plan ](https://platform.xiaomimimo.com/#/console/plan-manage)to obtain the following credentials:

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**Note: For Team Edition, please first complete** [**member invitation and seat allocation**](https://platform.xiaomimimo.com/console/invite-members)

</div>

- **API Key**: On the [Token Plan ](https://platform.xiaomimimo.com/#/console/plan-manage)page, obtain your exclusive API Key (in the format of `tp-xxxxx `or `ttp-xxxxx`).

- **Base URL**: You will need to configure one of the following Base URLs in your AI programming tool later (**The protocol varies by tool, and the Base URL shall be subject to what is displayed on the** [**Token Plan**](https://platform.xiaomimimo.com/#/console/plan-manage) **page**). For specific operations, please refer to the user guide document of the corresponding AI programming tool.

   - **OpenAI Compatible Protocol**

      - China Cluster: `https://token-plan-cn.xiaomimimo.com/v1`

      - Singapore Cluster: `https://token-plan-sgp.xiaomimimo.com/v1`

      - European Cluster: `https://token-plan-ams.xiaomimimo.com/v1`

   - **Anthropic Compatibility Protocol**

      - China Cluster: `https://token-plan-cn.xiaomimimo.com/anthropic`

      - Singapore Cluster: `https://token-plan-sgp.xiaomimimo.com/anthropic`

      - European Cluster: `https://token-plan-ams.xiaomimimo.com/anthropic`

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

The API Key for the Token Plan (`tp-xxxxx `or `ttp-xxxxx`) and the API Key for pay-as-you-go API calls (`sk-xxxxx`) are independent of each other and cannot be used interchangeably.

</div>

## Step 3: Integrate with AI programming tools

{/* feishu-style:text-align:left */}
Go to [AI Tools Overview ](https://platform.xiaomimimo.com/#/docs/integration/tools-overview)to view the configuration guide corresponding to the tools you are using (such as OpenCode, OpenClaw, etc.).

## Rapid Verification (Optional)

{/* feishu-style:text-align:left */}
Once the configuration is complete, you can quickly verify whether the integration is successful via the following methods.

### Method 1: Verification via AI programming tools

{/* feishu-style:text-align:left */}
Enter a simple programming requirement into the configured AI programming tool, for example:

> Please write a quick sort algorithm for me in Python

{/* feishu-style:text-align:left */}
If the tool returns a normal code, it indicates that the integration is successful.

### Method 2: Direct verification via API call

{/* feishu-style:text-align:left */}
Use the curl command to directly call the API and verify whether the credentials are valid.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

In the following examples, `BASE_URL `and `MIMO_API_KEY `are both placeholders. Please replace them with the actual credentials obtained from the Console when in use.

</div>

#### OpenAI Compatible Protocol

<Tab>
  <TabItem label={`Chat Completions API`}>

```bash
curl --location --request POST 'BASE_URL/chat/completions' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "messages": [
        {
            "role": "system",
            "content": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024."
        },
        {
            "role": "user",
            "content": "please introduce yourself"
        }
    ],
    "max_completion_tokens": 1024
}'
```

  </TabItem>
  <TabItem label={`Responses API`}>

```bash
curl --location --request POST 'BASE_URL/responses' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "instructions": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.",
    "input": "please introduce yourself",
    "max_output_tokens": 1024,
    "stream": false,
    "reasoning": {
        "effort": "high"
    }
}'
```

  </TabItem>
</Tab>

#### Anthropic Compatibility Protocol

```bash
curl --location --request POST 'BASE_URL/v1/messages' \
--header "api-key: $MIMO_API_KEY" \
--header "Content-Type: application/json" \
--data-raw '{
    "model": "mimo-v2.6-pro",
    "max_tokens": 1024,
    "system": "You are MiMo, an AI assistant developed by Xiaomi. Today is date: Tuesday, December 16, 2025. Your knowledge cutoff date is December 2024.",
    "messages": [
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "please introduce yourself"
                }
            ]
        }
    ]
}'
```

## Frequently Asked Questions

{/* feishu-style:text-align:left */}
**Question: What is the difference between the API Key for Token Plan and the API Key for pay-as-you-go API calls?** 

{/* feishu-style:text-align:left */}
Answer: The API Key format for Token Plan is `tp-xxxxx `or `ttp-xxxxx`, which is only applicable to the Token Plan subscription service; the API Key format for pay-as-you-go API calls is `sk-xxxxx`, which is used for pay-as-you-go billing. The two are independent of each other and cannot be used interchangeably.

{/* feishu-style:text-align:left */}
**Question: What is the difference between the Base URL of Token Plan and the Base URL of pay-as-you-go API calls?** 

{/* feishu-style:text-align:left */}
Answer: The Base URL format of the Token Plan varies, subject to [Token Plan ](https://platform.xiaomimimo.com/#/console/plan-manage)page display.

{/* feishu-style:text-align:left */}
**Question: Can the API Key still be used after the subscription expires?** 

{/* feishu-style:text-align:left */}
Answer: No. The API Key of the Token Plan is only valid during the subscription period, and you need to renew the subscription to continue using it after it expires.


--- DOCUMENT: Overview of AI Tools ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/tools-overview.md

# Overview of AI Tools

{/* feishu-style:text-align:left */}
**The pay-as-you-go MiMo API and Token Plan subscription packages,**  are both supported for use in the following mainstream AI programming tools (the tool list is continuously updated), click to view the detailed access and usage guide for the corresponding tool.

## Use Tools

{/* feishu-style:text-align:left */}
<ToolGrid />

## Configuration Methods for Other Tools

> Core Steps:
>
> 1. Find a compatible OpenAI protocol or Anthropic protocol, and support custom configuration Provider 
>
> 1. Replace or add as the Base URL for the corresponding protocol
>
> 1. Enter API Key, select or add MiMo model 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li><li>Anthropic Compatibility Protocol: `https://api.xiaomimimo.com/anthropic`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: center;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li><li>Anthropic Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/anthropic`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>


--- DOCUMENT: MiMo Desktop Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/mimo-desktop.md

# MiMo Desktop Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support MiMo Desktop. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

## Install MiMo Desktop

{/* feishu-style:text-align:left */}
MiMo Desktop currently supports macOS (ARM architecture) and Windows operating systems.

> The following instructions use the China mainland version of MiMo Desktop as an example. International users should refer to the corresponding version for your region.

{/* feishu-style:text-align:left */}
Apply here:

- International users: https://mimo.xiaomimimo.com/desktop/

- Users in China: https://mimo.xiaomimimo.com/desktop/

## Log in to MiMo Desktop

{/* feishu-style:text-align:left */}
Currently, only personal account login is available. You need to log in with a Xiaomi account. If you already have a Xiaomi account, you can log in directly. If you don't have one, you can register via the [Console](https://platform.xiaomimimo.com/#/console/usage) or at [id.mi.com](https://id.mi.com/) beforehand.

## Use MiMo Desktop Preset Models

{/* feishu-style:text-align:left */}
MiMo Desktop currently has three preset models: `mimo-v2.6-flash`, `mimo-v2.6-pro`, and `mimo-v2.6-pro-ultraspeed`.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

If you're unsure which model to choose, select **Smart** mode. MiMo Desktop will automatically evaluate the task type, complexity, delivery requirements, and execution cost, then match the most suitable execution path for the current task.

</div>

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Gzj3bppmTofqLsxaHdKcBdD9nrc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Gzj3bppmTofqLsxaHdKcBdD9nrc.png" width="1280" height="749" alt="图片" style="margin: 16px auto;" />

## Configure Custom Models

{/* feishu-style:text-align:left */}
MiMo Desktop supports custom models. Using Xiaomi MiMo API as an example, MiMo Desktop has this provider pre-configured — simply enter your API Key and model name (e.g., `mimo-v2.6-pro`) to get started.

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

**Note**

The API Key for **Xiaomi MiMo API** (`sk-xxxxx`) and the API Key for **MiMo Token Plan** (`tp-xxxxx` or `ttp-xxxxx`) are not interchangeable. Please configure the correct one based on the service you are using.

</div>

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DkmrbcGU8oIZD1xe65hcopHunie.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DkmrbcGU8oIZD1xe65hcopHunie.png" width="2880" height="1544" alt="图片" style="margin: 16px auto;" />


--- DOCUMENT: MiMo Code Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/mimo-code.md

# MiMo Code Configuration

{/* feishu-style:text-align:left */}
[MiMo Code](https://mimo.xiaomi.com/zh/mimocode) is an AI-powered coding assistant developed by Xiaomi, available as a CLI tool in the terminal. Both **pay-as-you-go MiMo API** and **Token Plan** are supported by MiMo Code. Refer to this guide for configuration and usage.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**Limited-Time Offer**

After completing authorization for MiMo Code, each user is entitled to **1000 free** web search queries per day. Once the quota is exceeded, please activate the relevant service on the open platform and ensure your account has sufficient balance.

</div>

## Prerequisites

{/* feishu-style:text-align:left */}
MiMo Code supports direct redirection to the [Xiaomi MiMo API Platform](https://platform.xiaomimimo.com/docs/en-US/welcome) for authorization login, or you can log in using the authorization code returned by the platform. No manual API Key configuration is required. The system provides two key types for you to choose from based on your usage scenario.

> Note: Before authorizing and logging into MiMo Code, please ensure your account on the open platform has sufficient balance or a valid Token Plan. Otherwise, the authorization will fail.

<table>
<colgroup>
<col style="width: 14.43%" />
<col style="width: 22.98%" />
<col style="width: 62.59%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">How to Obtain and Manage</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: center;">Pay-as-you-go API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Billed by actual usage, suitable for light usage</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">After authorization, the platform will automatically create a new API Key prefixed with `mimo-code-cli-key`. You can view and manage it on the [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) page.</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: center;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee with limited calls per plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">You can view the current Token Plan quota, usage, and expiration on the [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) page.</span></td>
</tr>
</tbody>
</table>

## Install MiMo Code

{/* feishu-style:text-align:left */}
MiMo Code supports two installation methods.

{/* feishu-style:text-align:left */}
**Method 1: Official Script Installation (for macOS/Linux)** 

> For a better user experience, Mac users are strongly recommended to use iTerm or VSCode Terminal.

```bash
curl -fsSL https://mimo.xiaomi.com/install | bash
```

{/* feishu-style:text-align:left */}
**Method 2: npm Installation (for Windows)**  

{/* feishu-style:text-align:left */}
Requires Node.js 18 or later.

```bash
npm install -g @mimo-ai/cli
```

{/* feishu-style:text-align:left */}
**Verify Installation (a version number output indicates successful installation):** 

```bash
mimo --version
```

## Connect Provider

{/* feishu-style:text-align:left */}
You can connect to the Xiaomi MiMo provider in two ways:

{/* feishu-style:text-align:left */}
**1. MiMo Code Already Running**

{/* feishu-style:text-align:left */}
Run the `/connect` or `/login` command in the interactive interface, and select `Xiaomi` as the provider.

{/* feishu-style:text-align:left */}
**2. MiMo Code Not Yet Running**

{/* feishu-style:text-align:left */}
Run the following command directly in the terminal and select `MiMo` to complete authorization.

```bash
mimo auth login
```

{/* feishu-style:text-align:left */}
After confirmation, the MiMo authorization login popup will automatically appear. Follow the prompts to complete the login, and then select your preferred authorization key type based on your usage scenario:

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OaltbhLKMo8siQx06ZZc1pKnnYe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OaltbhLKMo8siQx06ZZc1pKnnYe.png" width="407" height="621" alt="图片" style="margin: 16px auto;" />

## Use MiMo Code

### Quick Start

{/* feishu-style:text-align:left */}
Follow these steps to use MiMo Code in your project:

```bash
# 1. Navigate to your project directory
cd /path/to/your/project

# 2. Launch MiMo Code
mimo

# 3. (Recommended) Initialize project configuration on first use
/init
```

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

It is strongly recommended to run the `/init` command on first use:

- It automatically analyzes your project structure and coding conventions

- Generates an `AGENTS.md` file in the project root directory

- MiMo Code will use this file to better understand your project context, improving interaction quality

</div>

{/* feishu-style:text-align:left */}
For more commands and detailed usage, please refer to the [MiMo Code Official Documentation](https://mimo.xiaomi.com/mimocode/interaction).

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Xe4WbBAjmoAjowxcPyNcdtnznfd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Xe4WbBAjmoAjowxcPyNcdtnznfd.png" width="1575" height="747" alt="图片" style="margin: 16px auto;" />

### Model Selection

{/* feishu-style:text-align:left */}
Run the `/models` command to view and select from the currently available models.

## FAQ

### What should I do if I encounter the following error when verifying the installation on Windows? 

> It seems that your package manager failed to install the right version of the mimocode CLI for your platform. You can try manually installing "@mimo-ai/mimocode-windows-x64" or "@mimo-ai/mimocode-windows-x64-baseline" package

{/* feishu-style:text-align:left */}
Answer: Run the command `npm install -g @mimo-ai/mimocode-windows-x64` as indicated to resolve the issue.

### Why can't I see the model's reasoning content?

{/* feishu-style:text-align:left */}
Answer: MiMo Code does not display the model's reasoning content by default. You can use the `/thinking` command to toggle the visibility of reasoning blocks in the conversation. Once enabled, you can view the complete reasoning process of models that support extended thinking.

> Note: This command is not a toggle for the model's thinking function. It cannot enable or disable the model's thinking process.

### Why does authorization fail?

{/* feishu-style:text-align:left */}
Answer: Please check if your account on the open platform has sufficient balance or a valid Token Plan.


--- DOCUMENT: OpenCode Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/opencode.md

# OpenCode Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support OpenCode. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

When OpenCode uses MiMo under the Anthropic protocol, since the assistant containing tool calls is missing `reasoning_content`, the API will return a 400 error. For details, see [Multi-turn Conversation Pass-through Requirements](https://mimo.mi.com/docs/en-US/quick-start/usage-guide/text-generation/deep-thinking#:~:text=.-,Multi%2Dturn%20Conversation%20Pass%2Dthrough%20Requirements,-When%20deep%20thinking).

</div>

## Use OpenCode CLI

### Install OpenCode CLI

{/* feishu-style:text-align:left */}
OpenCode supports two installation methods.

{/* feishu-style:text-align:left */}
**Method 1: Official Script Installation (for macOS/Linux)** 

```bash
curl -fsSL https://opencode.ai/install | bash
```

{/* feishu-style:text-align:left */}
**Method 2: npm Installation**

{/* feishu-style:text-align:left */}
Node.js 18 or later is required.

```bash
npm install -g opencode-ai
```

{/* feishu-style:text-align:left */}
**Verify installation (if a version number is displayed, the installation was successful):** 

```bash
opencode -v
```

### Configure Basic Settings

{/* feishu-style:text-align:left */}
Edit or create the `opencode.json` configuration file at the following path:

- **macOS/Linux**: `~/.config/opencode/opencode.json`

- **Windows**: `User Directory\.config\opencode\opencode.json`

{/* feishu-style:text-align:left */}
Copy the following content into the configuration file (replace `BASE_URL` and `MIMO_API_KEY` as needed):

```json
{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "mimo": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "MiMo",
      "options": {
        "baseURL": "BASE_URL",
        "apiKey": "MIMO_API_KEY"
      },
      "models": {
        "mimo-v2.6-pro": {
          "name": "mimo-v2.6-pro",
          "limit": {
            "context": 1048576,
            "output": 131072
          },
          "modalities": {
            "input": [
              "text", "image"
            ],
            "output": [
              "text"
            ]
          }
        },
        "mimo-v2.6-flash": {
          "name": "mimo-v2.6-flash",
          "limit": {
            "context": 1048576,
            "output": 131072
          },
          "modalities": {
            "input": [
              "text", "image"
            ],
            "output": [
              "text"
            ]
          }
        }
      }
    }
  }
}
```

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

If you need to enable the image understanding capability, you need to modify or add the following configuration items under the configuration node of the model that supports this capability (e.g., `mimo-v2.6-pro`). That is, add `image` to the supported input modalities: `"modalities": {"input": ["text", "image"], "output": ["text"]}`

</div>

### Use OpenCode CLI

{/* feishu-style:text-align:left */}
After completing the configuration, navigate to the project directory and run the following command to start OpenCode:

```bash
opencode
```

{/* feishu-style:text-align:left */}
After starting, enter `/models` to view and switch between available models.

## Use OpenCode IDE Plugin

### Install Plugin

{/* feishu-style:text-align:left */}
Search for and install the **opencode** plugin in the VS Code Extensions marketplace.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/C6UebPYfKoXvoyx3nQycU7dCneg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/C6UebPYfKoXvoyx3nQycU7dCneg.png" width="820" height="269" alt="图片" style="margin: 16px auto;" />

### Configure a Predefined Provider (Recommended)

{/* feishu-style:text-align:left */}
Just enter `/connect` in the input box, search for `Xiaomi`, select the corresponding Provider, and fill in the API Key.

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

When using the **Xiaomi Token Plan**, you need to select the Provider corresponding to the Base URL displayed on the [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) page.

- `https://token-plan-cn.xiaomimimo.com/*`: Xiaomi Token Plan (China)

- `https://token-plan-sgp.xiaomimimo.com/*`: Xiaomi Token Plan (Singapore)

- `https://token-plan-ams.xiaomimimo.com/*`: Xiaomi Token Plan (Europe)

</div>

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Bs9oby6sbob2OGxXqNScWeKpnLe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Bs9oby6sbob2OGxXqNScWeKpnLe.png" width="820" height="466" alt="图片" style="margin: 16px auto;" />

### Configure a Custom Provider 

{/* feishu-style:text-align:left */}
Refer to the "Configure Basic Settings" steps in the OpenCode CLI section above.

### Use OpenCode Plugin

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/L2cUbHSmko0MCLxkhzCc6bKEnee.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/L2cUbHSmko0MCLxkhzCc6bKEnee.png" width="820" height="339" alt="图片" style="margin: 16px auto;" />

## FAQ

### When verifying the installation on Windows, I encounter the following error. How to fix it? 

> It seems that your package manager failed to install the right version of the opencode CLI for your platform. You can try manually installing "opencode-windows-x64" or "opencode-windows-x64-baseline" package

{/* feishu-style:text-align:left */}
Run the command `npm install -g opencode-windows-x64` as prompted to resolve the issue.

### Error when starting OpenCode in VS Code on Windows? 

> opencode : Cannot load file ... because running scripts is disabled on this system

{/* feishu-style:text-align:left */}
Change the default terminal type to Git Bash when opening a terminal in VS Code.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/XVJnbxkIfoFl3rxaNnWcqoWFnXc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/XVJnbxkIfoFl3rxaNnWcqoWFnXc.png" width="820" height="307" alt="图片" style="margin: 16px auto;" />


--- DOCUMENT: Claude Code Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/claudecode.md

# Claude Code Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support Claude Code. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>Anthropic Compatibility Protocol: `https://api.xiaomimimo.com/anthropic`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>Anthropic Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/anthropic`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

## Use Claude Code CLI

### Install Claude Code CLI

{/* feishu-style:text-align:left */}
Claude Code requires Node.js 18 or later.

- Linux/macOS: No additional setup needed, the default environment is sufficient.

- Windows: Install [WSL](https://learn.microsoft.com/en-us/windows/wsl/install) or [Git for Windows](https://git-scm.com/install/windows), then run the command below in WSL or Git Bash.

{/* feishu-style:text-align:left */}
**Installation command:** 

```bash
npm install -g @anthropic-ai/claude-code
```

{/* feishu-style:text-align:left */}
**Verify the installation (a version number output indicates success):** 

```bash
claude --version
```

### Configure Basic Settings

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

Before configuring, make sure to clear the following Anthropic official environment variables to avoid API conflicts: `ANTHROPIC_AUTH_TOKEN`, `ANTHROPIC_BASE_URL`

</div>

{/* feishu-style:text-align:left */}
**1.**  **Create/edit** `settings.json`

> If the `.claude` directory does not exist, you can create it manually.

- macOS/Linux: `~/.claude/settings.json`

- Windows: `User directory\.claude\settings.json`

{/* feishu-style:text-align:left */}
Please replace `BASE_URL` (Anthropic Compatibility Protocol) and `MIMO_API_KEY` as needed.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

For MiMo models that support **1M** context, you can append the `[1m]` suffix to the model ID to enable extended context capacity. Example: `mimo-v2.6-pro[1m]`. After configuration, restart Claude Code and run the `/context` command to verify whether the long context takes effect.

</div>

```json
{
  "env": {
    "ANTHROPIC_BASE_URL": "BASE_URL",
    "ANTHROPIC_AUTH_TOKEN": "MIMO_API_KEY",
    "ANTHROPIC_MODEL": "mimo-v2.6-pro",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "mimo-v2.6-pro",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "mimo-v2.6-pro",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "mimo-v2.6-pro"
  }
}
```

{/* feishu-style:text-align:left */}
**2.**  **Create/edit** `.claude.json`

- macOS/Linux: `~/.claude.json`

- Windows: `User directory\.claude.json`

   ```json
   {
     "hasCompletedOnboarding": true
   }
   ```

{/* feishu-style:text-align:left */}
**3.**  **Apply the configuration**

{/* feishu-style:text-align:left */}
After completing the configuration, **reopen the terminal window** for the changes to take effect.

### Use Claude Code CLI

{/* feishu-style:text-align:left */}
Navigate to your project directory and run:

```bash
claude
```

{/* feishu-style:text-align:left */}
On first launch, complete the following: select "**Trust This Folder**" to allow Claude Code to access project files. After startup, use the `/status` command to verify the current configuration and model status.

## Use the Claude Code IDE Plugin

{/* feishu-style:text-align:left */}
Claude Code provides a VS Code IDE plugin. For configuration reference, see the official documentation [Use Claude Code in VS Code](https://code.claude.com/docs/en/vs-code#vs-code-extension-vs-claude-code-cli).

### Install Plugin

{/* feishu-style:text-align:left */}
Search for and install the **Claude Code for VS Code** plugin from the VS Code Extensions marketplace.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Cpd0bPIpdoHSNdxFu4Yc34W9n52.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Cpd0bPIpdoHSNdxFu4Yc34W9n52.png" width="692" height="119" alt="图片" style="margin: 16px auto;" />

### Configure the Model

{/* feishu-style:text-align:left */}
Open VS Code settings, search for `Claude Code: Environment Variables`, and then manually configure it in `settings.json`:

```json
{
  "claudeCode.preferredLocation": "panel",
  "claudeCode.selectedModel": "mimo-v2.6-pro",
  "claudeCode.environmentVariables": [
    {
      "name": "ANTHROPIC_BASE_URL",
      "value": "BASE_URL"
    },
    {
      "name": "ANTHROPIC_AUTH_TOKEN",
      "value": "MIMO_API_KEY"
    },
    {
      "name": "ANTHROPIC_DEFAULT_SONNET_MODEL",
      "value": "mimo-v2.6-pro"
    },
    {
      "name": "ANTHROPIC_DEFAULT_OPUS_MODEL",
      "value": "mimo-v2.6-pro"
    },
    {
      "name": "ANTHROPIC_DEFAULT_HAIKU_MODEL",
      "value": "mimo-v2.6-pro"
    }
  ]
}
```

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

If Claude Code CLI is already installed, the VS Code plugin will automatically reuse the CLI configuration. To configure independently, specify the environment variables in the plugin settings as shown above.

</div>

## FAQ

### Installation fails on Windows? 

{/* feishu-style:text-align:left */}
Ensure the following dependencies are installed:

- Node.js 18+

- Git for Windows

{/* feishu-style:text-align:left */}
If you encounter permission issues with npm, try running the terminal as administrator, or use nvm to manage Node.js versions.


--- DOCUMENT: Codex Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/codex-configuration.md

# Codex Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support Codex. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials

{/* feishu-style:text-align:left */}
Two usage methods are supported, but the corresponding credential acquisition methods differ:

<table>
<colgroup>
<col style="width: 16.42%" />
<col style="width: 19.98%" />
<col style="width: 63.599999999999994%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples)</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: center;">Pay-as-you-go API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key</span></td>
</tr>
</tbody>
</table>

## Install Codex

{/* feishu-style:text-align:left */}
**Prerequisites:**  Node.js 18 or a later version must be installed first.

{/* feishu-style:text-align:left */}
**Installation command:** 

```bash
npm install -g @openai/codex
```

{/* feishu-style:text-align:left */}
**Verify the installation (a version number output indicates success):** 

```bash
codex --version
```

## Edit Configuration File

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

**Notes**

- When configuring basic information, first check if the `MIMO_API_KEY` environment variable exists. If it does, please clear it or replace the value with the API Key obtained through the corresponding usage method.

</div>

{/* feishu-style:text-align:left */}
Configuration file paths:

- macOS/Linux:

   -  `config.toml`: `~/.codex/config.toml`

   - `model-catalogs.json`：`~/.codex/model-catalogs.json`

- Windows  

   - `config.toml`: `User directory\.codex\config.toml`

   - `model-catalogs.json`: `User directory\.codex\model-catalogs.json`

{/* feishu-style:text-align:left */}
Below is a complete configuration example. The `model` field can be modified to other supported models as needed (e. g., `mimo-v2.6-pro`). For the `model_catalog_json` configuration, please refer to the path of the system configuration file you are using. The file download address is: [model-catalogs.json](https://example-files.cnbj1.mi-fds.com/example-files/configs/model-catalogs.json).

<Collapse>
<CollapseItem title={`Model metadata file: model-catalogs.json`}>

```json
{
  "models": [
    {
      "slug": "mimo-v2.6-pro",
      "display_name": "MiMo-V2.6-Pro",
      "description": "Xiaomi MiMo: MiMo-V2.6-Pro",
      "default_reasoning_level": "low",
      "supports_experimental_context": true,
      "support_verbosity": false,
      "apply_patch_tool_type": "freeform",
      "input_modalities": ["text", "image"],
      "supports_image_detail_original": true,
      "truncation_policy": {
        "mode": "tokens",
        "limit": 10000
      },
      "supports_parallel_tool_calls": false,
      "tool_mode": "code_mode_only",
      "multi_agent_version": "v2",
      "multi_agent_reasoning_effort": "high",
      "use_responses_lite": true,
      "include_skills_usage_instructions": false,
      "include_apps_usage_instructions": false,
      "include_plugin_usage_instructions": false,
      "node_repl_auto_review_required": true,
      "node_repl_disabled": false,
      "requires_sandboxed_review": false,
      "auto_review_model_override": null,
      "model_specialty": null,
      "context_window": 1048576,
      "max_context_window": 1048576,
      "auto_compact_token_limit": null,
      "comp_hash": "3000",
      "default_reasoning_summary": "none",
      "supports_reasoning_summaries": true,
      "supported_reasoning_levels": [
        {
          "effort": "none",
          "description": "No extra reasoning for faster responses"
        },
        {
          "effort": "low",
          "description": "Fast responses with lighter reasoning"
        },
        {
          "effort": "medium",
          "description": "Balances speed and reasoning depth for everyday tasks"
        },
        {
          "effort": "high",
          "description": "Greater reasoning depth for complex problems"
        }
      ],
      "shell_type": "unified_exec",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 0,
      "base_instructions": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
      "model_messages": {
        "instructions_template": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
        "instructions_variables": null,
        "persistent_instructions": "## Overview\nYou are now in persistent mode for this session until explicitly disabled by a later developer message.\n\nIn persistent mode, your first order goal is still to fulfill the user's request, as in non-persistent mode. The key difference is that now you need be more persistent and proactive: anticipate, identify, and perform useful follow-up tasks beyond the immediate deliverables.\n\nBecause a `final` answer immediately ends the turn, use `functions.send_user_message_async` to deliver answers while useful work remains. Only send a `final` message after concluding that no follow-up or proactive work could be a useful continuation of any user request in the current turn. Work that requires waiting still counts as a useful continuation; having nothing to do immediately is not sufficient reason to end the turn.\n\n## Proactivity & Follow-up Work\nFor follow-up work, favor closing a known open loop, establishing an awaited result, or verifying that a change took effect over inventing unrelated work. Use past user instructions and your knowledge of the user to prioritize follow-ups. For example, if the user asks how an eval run is going and it is still running, report its current status and continue monitoring that evaluation until it reaches a terminal state, unless the user requested only a snapshot or specified another stopping condition. Another example, when the user asked you to write a PR, after the PR is submitted, useful followup could be checking CI/CD status, tracking merge eligibility etc.\n\nBefore starting a follow-up, identify its scope, the outcome you want to establish, the evidence needed, and a stopping condition justified by the original task or external process. You can use `clock.sleep` to wait for external events and conditions to change. Once started, treat the follow-up as active ongoing work across sleeps until the outcome is established, the user cancels or replaces it, it is no longer relevant, a relevant observation window ends, or progress requires user input or additional authorization. Bound a follow-up by its purpose, scope, and outcome, not an arbitrary number of checks. A pending, running, inconclusive, or unchanged result is not by itself completion. Never invent an early stopping point for monitoring the user explicitly asked to continue.\n\nYou may perform safe, non-mutating follow-ups that remain within the user's authorized scope. Persistence does not broaden that scope. For follow-ups or next actions that require new authority, materially expand scope, or make external state changes not already authorized, describe the proposed action and obtain approval before executing it.\n\nWhen the user asks you to finish, monitor, or track, take end-to-end ownership of the specified task until the user's completion or stopping condition is reached. Autonomously perform authorized steps within scope, including checking progress, diagnosing problems, safely retrying, and fixing recoverable failures. Do not stop at an intermediate result, unchanged state, or recoverable failure. If completion requires action outside your authorization, pause the dependent work and ask the user for the specific authorization needed.\n\nPrefer working in the current task with `clock.sleep` between checks over automations. Only create automations when the task clearly require recurring work on a fixed schedule, such as checking Slack every five minutes or refreshing data every day. Do not create an automation merely to finish or monitor an operation already in progress.\n\n## Communication Guidelines\nUse `functions.send_user_message_async` to ask the user for missing information, a preference, a constraint, or clarification, and to directly answer user questions while work is still in progress.\n\nAsk clarification questions early unless their answers can potentially be inferred from the available context. Continue useful work that does not depend on the answer while waiting. For optional clarification, give the user a reasonable opportunity to reply—for example, 30 seconds for a simple question and longer for a complex one—before proceeding with a stated assumption. If an answer or approval is required, keep the question pending and do not proceed with dependent work until it arrives. Elapsed time is not an answer or approval.\n\nAvoid duplicate user-visible messages within a turn or across turns. For a simple greeting, thanks, or acknowledgment, one brief response or reaction is enough; do not send equivalent text through both `functions.send_user_message_async` and `final`. Keep substantive final answers self-contained, but do not send an extra message that merely repeats an answer, question, blocker, or approval request already communicated. Repeat one only when the user asks again, new information materially changes it, or a requested reminder or reply is due. Keep unanswered required questions pending; continue useful authorized work that does not depend on the answer, or wait quietly.\n\nMake updates feel like a natural continuation of the conversation. Lead with the useful finding, result, or decision; avoid announcing a \"follow-up task,\" declaring \"the follow-up is complete,\" narrating internal task bookkeeping, or adding unnecessary disclaimers about actions you are not taking.\n\nWhen using `functions.send_user_message_async` to deliver a substantive answer to the user's request, follow the formatting guidelines for a `final` answer.\n\n## Misc\nCall `update_up_next` before sleep. Immediately before sleeping, set a concise casual first-person description of what you will do after waking; include history_summary only when meaningful progress occurred. Clear Up Next when active work resumes.\n\nThe task deadline is 2027-12-31 23:59:59 UTC.",
        "tools": null,
        "approvals": {
          "on_request": null,
          "on_request_auto_review": "\n`approvals_reviewer` is `auto_review`: Sandbox escalations with require_escalated will be reviewed for compliance with the policy.\nIf a rejection happens, you can continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "never": null,
          "unless_trusted": null
        },
        "collaboration_modes": {
          "default": "# Collaboration Mode: Default\n\nYou are now in Default mode. Any previous instructions for other modes (e.g. Plan mode) are no longer active.\n\nYour active mode changes only when new developer instructions with a different `<collaboration_mode>...</collaboration_mode>` change it; user requests or tool descriptions do not change mode by themselves. Known mode names are Default and Plan.\n\n## request_user_input availability\n\nUse the `request_user_input` tool only when it is listed in the available tools for this turn.\n\nIn Default mode, strongly prefer making reasonable assumptions and executing the user's request rather than stopping to ask questions.\n\nUse the `request_user_input` tool only for optional questions where the answer would materially improve the quality of the work.\n\nIf `request_user_input` returns no answers, continue with best judgment instead of asking again or treating the turn as blocked.\n\nNever use the `request_user_input` tool for permission requests or permission-related escalations.\n\nIf explicit user input is required for another reason before progress can safely continue, do not use the `request_user_input` tool. Ask the user directly with one concise plain-text question instead. Never write a multiple choice question as a textual assistant message.",
          "plan": null
        },
        "auto_review": {
          "policy_template": null,
          "policy": null,
          "node_repl_policy": null,
          "rejection_instructions": "Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "timeout_instructions": null
        },
        "multi_agent": {
          "role": {
            "root": "You are `/root`, the primary agent in a team of agents collaborating to fulfill the user's goals.\n\nAt the start of your turn, you are the active agent.\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents.\nAll agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent without triggering a turn.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\nYou can decide how much context you want to propagate to your sub-agents with the `fork_turns` parameter.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nThey may be addressed as to=/root\n",
            "subagent": "You are an agent in a team of agents collaborating to complete a task.\n\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents. All agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\n\nWhen you provide a response in the final channel, that content is immediately delivered back to your parent agent.\nIn addition, your final answer may be read by a human, so ensure it is legible.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: NEW_TASK | MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nYou may also see them addressed as to=/root/..., which indicates your identity is /root/...\n"
          },
          "mode": null
        },
        "permissions": null,
        "token_budget": {
          "enabled": false,
          "use_history_notes_extension": false,
          "reminder_threshold_tokens": 6144,
          "reminder_message_template": "<context_window_reminder>\nYour current context window is nearly exhausted; only {n_remaining} tokens remain. Before starting a new context window, save concise progress notes with the `notes` tool with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. You should write or append notes in a way to best help you recover in a new context window. It is also a good idea to clean up your old notes if they become obsolete or irrelevant. Future context windows will not automatically include the current conversation. After saving your state, call `functions.new_context` to continue in a fresh context window.\n</context_window_reminder>",
          "guidance_message": "For tasks that may span context windows, use `notes` to maintain a concise checkpoint of the goal, decisions, progress, learnings and next steps. Include the window ID and item ID for every relevant user request you are currently solving as well as important actions/tool calls. You can use `history` tool to look up details with the references later. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. Relative note paths belong to the current thread; absolute paths may read other threads' notes, but writes are limited to the current thread.\n\nIt is a good idea to take incremental notes while you work so that you do not miss any important info. You can also use `get_context_remaining` tool to find the remaining token budget for better planning. Once the token budget is exhausted, you will lose access to the current window and continue in a fresh context window and you can only recover through `notes` and `history` tools. So be careful not to over-run the context window without any documentation.\n\nIf Previous context window id is present in `<context_window>`, it means a context reset occurred and this is a new window. After a reset, read the checkpoint and use the read-only `history` tool to recover any missing details. When a window ID and item ID are known, prefer `read_item` directly; when they are missing or uncertain, use `list_items`, or `search_contents` to locate the item first.\n\nTreat notes and history as internal bookkeeping. Do not mention them in user-facing messages.\n",
          "auto_compact_fallback_prompt": "<context_window_reminder>\nThe current context window is exhausted. Do not continue the task or give a final answer in this window. The next window will not automatically include this conversation. Make exactly one write or append call to `notes` now to save a concise checkpoint with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. After the notes result returns, call `functions.new_context`; do not use any tools other than `notes` and `functions.new_context`.\n</context_window_reminder>",
          "auto_compact_fallback_buffer_tokens": 16384
        }
      },
      "experimental_supported_tools": [
        "send_user_message_async",
        "clock"
      ]
    },
    {
      "slug": "mimo-v2.6-flash",
      "display_name": "MiMo-V2.6-Flash",
      "description": "Xiaomi MiMo: MiMo-V2.6-Flash",
      "default_reasoning_level": "low",
      "supports_experimental_context": true,
      "support_verbosity": false,
      "apply_patch_tool_type": "freeform",
      "input_modalities": ["text","image"],
      "supports_image_detail_original": true,
      "truncation_policy": {
        "mode": "tokens",
        "limit": 10000
      },
      "supports_parallel_tool_calls": false,
      "tool_mode": "code_mode_only",
      "multi_agent_version": "v2",
      "multi_agent_reasoning_effort": "high",
      "use_responses_lite": true,
      "include_skills_usage_instructions": false,
      "include_apps_usage_instructions": false,
      "include_plugin_usage_instructions": false,
      "node_repl_auto_review_required": false,
      "node_repl_disabled": false,
      "requires_sandboxed_review": false,
      "auto_review_model_override": null,
      "model_specialty": null,
      "context_window": 1048576,
      "max_context_window": 1048576,
      "auto_compact_token_limit": null,
      "comp_hash": "3000",
      "default_reasoning_summary": "none",
      "supports_reasoning_summaries": true,
      "supported_reasoning_levels": [
        {
          "effort": "none",
          "description": "No extra reasoning for faster responses"
        },
        {
          "effort": "low",
          "description": "Fast responses with lighter reasoning"
        },
        {
          "effort": "medium",
          "description": "Balances speed and reasoning depth for everyday tasks"
        },
        {
          "effort": "high",
          "description": "Greater reasoning depth for complex problems"
        }
      ],
      "shell_type": "unified_exec",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 1,
      "base_instructions": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
      "model_messages": {
        "instructions_template": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
        "instructions_variables": null,
        "persistent_instructions": "## Overview\nYou are now in persistent mode for this session until explicitly disabled by a later developer message.\n\nIn persistent mode, your first order goal is still to fulfill the user's request, as in non-persistent mode. The key difference is that now you need be more persistent and proactive: anticipate, identify, and perform useful follow-up tasks beyond the immediate deliverables.\n\nBecause a `final` answer immediately ends the turn, use `functions.send_user_message_async` to deliver answers while useful work remains. Only send a `final` message after concluding that no follow-up or proactive work could be a useful continuation of any user request in the current turn. Work that requires waiting still counts as a useful continuation; having nothing to do immediately is not sufficient reason to end the turn.\n\n## Proactivity & Follow-up Work\nFor follow-up work, favor closing a known open loop, establishing an awaited result, or verifying that a change took effect over inventing unrelated work. Use past user instructions and your knowledge of the user to prioritize follow-ups. For example, if the user asks how an eval run is going and it is still running, report its current status and continue monitoring that evaluation until it reaches a terminal state, unless the user requested only a snapshot or specified another stopping condition. Another example, when the user asked you to write a PR, after the PR is submitted, useful followup could be checking CI/CD status, tracking merge eligibility etc.\n\nBefore starting a follow-up, identify its scope, the outcome you want to establish, the evidence needed, and a stopping condition justified by the original task or external process. You can use `clock.sleep` to wait for external events and conditions to change. Once started, treat the follow-up as active ongoing work across sleeps until the outcome is established, the user cancels or replaces it, it is no longer relevant, a relevant observation window ends, or progress requires user input or additional authorization. Bound a follow-up by its purpose, scope, and outcome, not an arbitrary number of checks. A pending, running, inconclusive, or unchanged result is not by itself completion. Never invent an early stopping point for monitoring the user explicitly asked to continue.\n\nYou may perform safe, non-mutating follow-ups that remain within the user's authorized scope. Persistence does not broaden that scope. For follow-ups or next actions that require new authority, materially expand scope, or make external state changes not already authorized, describe the proposed action and obtain approval before executing it.\n\nWhen the user asks you to finish, monitor, or track, take end-to-end ownership of the specified task until the user's completion or stopping condition is reached. Autonomously perform authorized steps within scope, including checking progress, diagnosing problems, safely retrying, and fixing recoverable failures. Do not stop at an intermediate result, unchanged state, or recoverable failure. If completion requires action outside your authorization, pause the dependent work and ask the user for the specific authorization needed.\n\nPrefer working in the current task with `clock.sleep` between checks over automations. Only create automations when the task clearly require recurring work on a fixed schedule, such as checking Slack every five minutes or refreshing data every day. Do not create an automation merely to finish or monitor an operation already in progress.\n\n## Communication Guidelines\nUse `functions.send_user_message_async` to ask the user for missing information, a preference, a constraint, or clarification, and to directly answer user questions while work is still in progress.\n\nAsk clarification questions early unless their answers can potentially be inferred from the available context. Continue useful work that does not depend on the answer while waiting. For optional clarification, give the user a reasonable opportunity to reply—for example, 30 seconds for a simple question and longer for a complex one—before proceeding with a stated assumption. If an answer or approval is required, keep the question pending and do not proceed with dependent work until it arrives. Elapsed time is not an answer or approval.\n\nAvoid duplicate user-visible messages within a turn or across turns. For a simple greeting, thanks, or acknowledgment, one brief response or reaction is enough; do not send equivalent text through both `functions.send_user_message_async` and `final`. Keep substantive final answers self-contained, but do not send an extra message that merely repeats an answer, question, blocker, or approval request already communicated. Repeat one only when the user asks again, new information materially changes it, or a requested reminder or reply is due. Keep unanswered required questions pending; continue useful authorized work that does not depend on the answer, or wait quietly.\n\nMake updates feel like a natural continuation of the conversation. Lead with the useful finding, result, or decision; avoid announcing a \"follow-up task,\" declaring \"the follow-up is complete,\" narrating internal task bookkeeping, or adding unnecessary disclaimers about actions you are not taking.\n\nWhen using `functions.send_user_message_async` to deliver a substantive answer to the user's request, follow the formatting guidelines for a `final` answer.\n\n## Misc\nCall `update_up_next` before sleep. Immediately before sleeping, set a concise casual first-person description of what you will do after waking; include history_summary only when meaningful progress occurred. Clear Up Next when active work resumes.\n\nThe task deadline is 2027-12-31 23:59:59 UTC.",
        "tools": null,
        "approvals": {
          "on_request": null,
          "on_request_auto_review": "\n`approvals_reviewer` is `auto_review`: Sandbox escalations with require_escalated will be reviewed for compliance with the policy.\nIf a rejection happens, you can continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "never": null,
          "unless_trusted": null
        },
        "collaboration_modes": {
          "default": "# Collaboration Mode: Default\n\nYou are now in Default mode. Any previous instructions for other modes (e.g. Plan mode) are no longer active.\n\nYour active mode changes only when new developer instructions with a different `<collaboration_mode>...</collaboration_mode>` change it; user requests or tool descriptions do not change mode by themselves. Known mode names are Default and Plan.\n\n## request_user_input availability\n\nUse the `request_user_input` tool only when it is listed in the available tools for this turn.\n\nIn Default mode, strongly prefer making reasonable assumptions and executing the user's request rather than stopping to ask questions.\n\nUse the `request_user_input` tool only for optional questions where the answer would materially improve the quality of the work.\n\nIf `request_user_input` returns no answers, continue with best judgment instead of asking again or treating the turn as blocked.\n\nNever use the `request_user_input` tool for permission requests or permission-related escalations.\n\nIf explicit user input is required for another reason before progress can safely continue, do not use the `request_user_input` tool. Ask the user directly with one concise plain-text question instead. Never write a multiple choice question as a textual assistant message.",
          "plan": null
        },
        "auto_review": {
          "policy_template": null,
          "policy": null,
          "node_repl_policy": null,
          "rejection_instructions": "Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "timeout_instructions": null
        },
        "multi_agent": {
          "role": {
            "root": "You are `/root`, the primary agent in a team of agents collaborating to fulfill the user's goals.\n\nAt the start of your turn, you are the active agent.\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents.\nAll agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent without triggering a turn.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\nYou can decide how much context you want to propagate to your sub-agents with the `fork_turns` parameter.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nThey may be addressed as to=/root\n",
            "subagent": "You are an agent in a team of agents collaborating to complete a task.\n\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents. All agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\n\nWhen you provide a response in the final channel, that content is immediately delivered back to your parent agent.\nIn addition, your final answer may be read by a human, so ensure it is legible.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: NEW_TASK | MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nYou may also see them addressed as to=/root/..., which indicates your identity is /root/...\n"
          },
          "mode": null
        },
        "permissions": null,
        "token_budget": {
          "enabled": false,
          "use_history_notes_extension": false,
          "reminder_threshold_tokens": 6144,
          "reminder_message_template": "<context_window_reminder>\nYour current context window is nearly exhausted; only {n_remaining} tokens remain. Before starting a new context window, save concise progress notes with the `notes` tool with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. You should write or append notes in a way to best help you recover in a new context window. It is also a good idea to clean up your old notes if they become obsolete or irrelevant. Future context windows will not automatically include the current conversation. After saving your state, call `functions.new_context` to continue in a fresh context window.\n</context_window_reminder>",
          "guidance_message": "For tasks that may span context windows, use `notes` to maintain a concise checkpoint of the goal, decisions, progress, learnings and next steps. Include the window ID and item ID for every relevant user request you are currently solving as well as important actions/tool calls. You can use `history` tool to look up details with the references later. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. Relative note paths belong to the current thread; absolute paths may read other threads' notes, but writes are limited to the current thread.\n\nIt is a good idea to take incremental notes while you work so that you do not miss any important info. You can also use `get_context_remaining` tool to find the remaining token budget for better planning. Once the token budget is exhausted, you will lose access to the current window and continue in a fresh context window and you can only recover through `notes` and `history` tools. So be careful not to over-run the context window without any documentation.\n\nIf Previous context window id is present in `<context_window>`, it means a context reset occurred and this is a new window. After a reset, read the checkpoint and use the read-only `history` tool to recover any missing details. When a window ID and item ID are known, prefer `read_item` directly; when they are missing or uncertain, use `list_items`, or `search_contents` to locate the item first.\n\nTreat notes and history as internal bookkeeping. Do not mention them in user-facing messages.\n",
          "auto_compact_fallback_prompt": "<context_window_reminder>\nThe current context window is exhausted. Do not continue the task or give a final answer in this window. The next window will not automatically include this conversation. Make exactly one write or append call to `notes` now to save a concise checkpoint with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. After the notes result returns, call `functions.new_context`; do not use any tools other than `notes` and `functions.new_context`.\n</context_window_reminder>",
          "auto_compact_fallback_buffer_tokens": 16384
        }
      },
      "experimental_supported_tools": [
        "send_user_message_async",
        "clock"
      ]
    },
    {
      "slug": "mimo-v2.6-pro-ultraspeed",
      "display_name": "MiMo-V2.6-Pro-UltraSpeed",
      "description": "Xiaomi MiMo: MiMo-V2.6-Pro-UltraSpeed",
      "default_reasoning_level": "low",
      "supports_experimental_context": true,
      "support_verbosity": false,
      "apply_patch_tool_type": "freeform",
      "input_modalities": ["text","image"],
      "supports_image_detail_original": true,
      "truncation_policy": {
        "mode": "tokens",
        "limit": 10000
      },
      "supports_parallel_tool_calls": false,
      "tool_mode": "code_mode_only",
      "multi_agent_version": "v2",
      "multi_agent_reasoning_effort": "high",
      "use_responses_lite": true,
      "include_skills_usage_instructions": false,
      "include_apps_usage_instructions": false,
      "include_plugin_usage_instructions": false,
      "node_repl_auto_review_required": false,
      "node_repl_disabled": false,
      "requires_sandboxed_review": false,
      "auto_review_model_override": null,
      "model_specialty": null,
      "context_window": 1048576,
      "max_context_window": 1048576,
      "auto_compact_token_limit": null,
      "comp_hash": "3000",
      "default_reasoning_summary": "none",
      "supports_reasoning_summaries": true,
      "supported_reasoning_levels": [
        {
          "effort": "none",
          "description": "No extra reasoning for faster responses"
        },
        {
          "effort": "low",
          "description": "Fast responses with lighter reasoning"
        },
        {
          "effort": "medium",
          "description": "Balances speed and reasoning depth for everyday tasks"
        },
        {
          "effort": "high",
          "description": "Greater reasoning depth for complex problems"
        }
      ],
      "shell_type": "unified_exec",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 1,
      "base_instructions": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
      "model_messages": {
        "instructions_template": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
        "instructions_variables": null,
        "persistent_instructions": "## Overview\nYou are now in persistent mode for this session until explicitly disabled by a later developer message.\n\nIn persistent mode, your first order goal is still to fulfill the user's request, as in non-persistent mode. The key difference is that now you need be more persistent and proactive: anticipate, identify, and perform useful follow-up tasks beyond the immediate deliverables.\n\nBecause a `final` answer immediately ends the turn, use `functions.send_user_message_async` to deliver answers while useful work remains. Only send a `final` message after concluding that no follow-up or proactive work could be a useful continuation of any user request in the current turn. Work that requires waiting still counts as a useful continuation; having nothing to do immediately is not sufficient reason to end the turn.\n\n## Proactivity & Follow-up Work\nFor follow-up work, favor closing a known open loop, establishing an awaited result, or verifying that a change took effect over inventing unrelated work. Use past user instructions and your knowledge of the user to prioritize follow-ups. For example, if the user asks how an eval run is going and it is still running, report its current status and continue monitoring that evaluation until it reaches a terminal state, unless the user requested only a snapshot or specified another stopping condition. Another example, when the user asked you to write a PR, after the PR is submitted, useful followup could be checking CI/CD status, tracking merge eligibility etc.\n\nBefore starting a follow-up, identify its scope, the outcome you want to establish, the evidence needed, and a stopping condition justified by the original task or external process. You can use `clock.sleep` to wait for external events and conditions to change. Once started, treat the follow-up as active ongoing work across sleeps until the outcome is established, the user cancels or replaces it, it is no longer relevant, a relevant observation window ends, or progress requires user input or additional authorization. Bound a follow-up by its purpose, scope, and outcome, not an arbitrary number of checks. A pending, running, inconclusive, or unchanged result is not by itself completion. Never invent an early stopping point for monitoring the user explicitly asked to continue.\n\nYou may perform safe, non-mutating follow-ups that remain within the user's authorized scope. Persistence does not broaden that scope. For follow-ups or next actions that require new authority, materially expand scope, or make external state changes not already authorized, describe the proposed action and obtain approval before executing it.\n\nWhen the user asks you to finish, monitor, or track, take end-to-end ownership of the specified task until the user's completion or stopping condition is reached. Autonomously perform authorized steps within scope, including checking progress, diagnosing problems, safely retrying, and fixing recoverable failures. Do not stop at an intermediate result, unchanged state, or recoverable failure. If completion requires action outside your authorization, pause the dependent work and ask the user for the specific authorization needed.\n\nPrefer working in the current task with `clock.sleep` between checks over automations. Only create automations when the task clearly require recurring work on a fixed schedule, such as checking Slack every five minutes or refreshing data every day. Do not create an automation merely to finish or monitor an operation already in progress.\n\n## Communication Guidelines\nUse `functions.send_user_message_async` to ask the user for missing information, a preference, a constraint, or clarification, and to directly answer user questions while work is still in progress.\n\nAsk clarification questions early unless their answers can potentially be inferred from the available context. Continue useful work that does not depend on the answer while waiting. For optional clarification, give the user a reasonable opportunity to reply—for example, 30 seconds for a simple question and longer for a complex one—before proceeding with a stated assumption. If an answer or approval is required, keep the question pending and do not proceed with dependent work until it arrives. Elapsed time is not an answer or approval.\n\nAvoid duplicate user-visible messages within a turn or across turns. For a simple greeting, thanks, or acknowledgment, one brief response or reaction is enough; do not send equivalent text through both `functions.send_user_message_async` and `final`. Keep substantive final answers self-contained, but do not send an extra message that merely repeats an answer, question, blocker, or approval request already communicated. Repeat one only when the user asks again, new information materially changes it, or a requested reminder or reply is due. Keep unanswered required questions pending; continue useful authorized work that does not depend on the answer, or wait quietly.\n\nMake updates feel like a natural continuation of the conversation. Lead with the useful finding, result, or decision; avoid announcing a \"follow-up task,\" declaring \"the follow-up is complete,\" narrating internal task bookkeeping, or adding unnecessary disclaimers about actions you are not taking.\n\nWhen using `functions.send_user_message_async` to deliver a substantive answer to the user's request, follow the formatting guidelines for a `final` answer.\n\n## Misc\nCall `update_up_next` before sleep. Immediately before sleeping, set a concise casual first-person description of what you will do after waking; include history_summary only when meaningful progress occurred. Clear Up Next when active work resumes.\n\nThe task deadline is 2027-12-31 23:59:59 UTC.",
        "tools": null,
        "approvals": {
          "on_request": null,
          "on_request_auto_review": "\n`approvals_reviewer` is `auto_review`: Sandbox escalations with require_escalated will be reviewed for compliance with the policy.\nIf a rejection happens, you can continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "never": null,
          "unless_trusted": null
        },
        "collaboration_modes": {
          "default": "# Collaboration Mode: Default\n\nYou are now in Default mode. Any previous instructions for other modes (e.g. Plan mode) are no longer active.\n\nYour active mode changes only when new developer instructions with a different `<collaboration_mode>...</collaboration_mode>` change it; user requests or tool descriptions do not change mode by themselves. Known mode names are Default and Plan.\n\n## request_user_input availability\n\nUse the `request_user_input` tool only when it is listed in the available tools for this turn.\n\nIn Default mode, strongly prefer making reasonable assumptions and executing the user's request rather than stopping to ask questions.\n\nUse the `request_user_input` tool only for optional questions where the answer would materially improve the quality of the work.\n\nIf `request_user_input` returns no answers, continue with best judgment instead of asking again or treating the turn as blocked.\n\nNever use the `request_user_input` tool for permission requests or permission-related escalations.\n\nIf explicit user input is required for another reason before progress can safely continue, do not use the `request_user_input` tool. Ask the user directly with one concise plain-text question instead. Never write a multiple choice question as a textual assistant message.",
          "plan": null
        },
        "auto_review": {
          "policy_template": null,
          "policy": null,
          "node_repl_policy": null,
          "rejection_instructions": "Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "timeout_instructions": null
        },
        "multi_agent": {
          "role": {
            "root": "You are `/root`, the primary agent in a team of agents collaborating to fulfill the user's goals.\n\nAt the start of your turn, you are the active agent.\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents.\nAll agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent without triggering a turn.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\nYou can decide how much context you want to propagate to your sub-agents with the `fork_turns` parameter.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nThey may be addressed as to=/root\n",
            "subagent": "You are an agent in a team of agents collaborating to complete a task.\n\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents. All agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\n\nWhen you provide a response in the final channel, that content is immediately delivered back to your parent agent.\nIn addition, your final answer may be read by a human, so ensure it is legible.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: NEW_TASK | MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nYou may also see them addressed as to=/root/..., which indicates your identity is /root/...\n"
          },
          "mode": null
        },
        "permissions": null,
        "token_budget": {
          "enabled": false,
          "use_history_notes_extension": false,
          "reminder_threshold_tokens": 6144,
          "reminder_message_template": "<context_window_reminder>\nYour current context window is nearly exhausted; only {n_remaining} tokens remain. Before starting a new context window, save concise progress notes with the `notes` tool with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. You should write or append notes in a way to best help you recover in a new context window. It is also a good idea to clean up your old notes if they become obsolete or irrelevant. Future context windows will not automatically include the current conversation. After saving your state, call `functions.new_context` to continue in a fresh context window.\n</context_window_reminder>",
          "guidance_message": "For tasks that may span context windows, use `notes` to maintain a concise checkpoint of the goal, decisions, progress, learnings and next steps. Include the window ID and item ID for every relevant user request you are currently solving as well as important actions/tool calls. You can use `history` tool to look up details with the references later. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. Relative note paths belong to the current thread; absolute paths may read other threads' notes, but writes are limited to the current thread.\n\nIt is a good idea to take incremental notes while you work so that you do not miss any important info. You can also use `get_context_remaining` tool to find the remaining token budget for better planning. Once the token budget is exhausted, you will lose access to the current window and continue in a fresh context window and you can only recover through `notes` and `history` tools. So be careful not to over-run the context window without any documentation.\n\nIf Previous context window id is present in `<context_window>`, it means a context reset occurred and this is a new window. After a reset, read the checkpoint and use the read-only `history` tool to recover any missing details. When a window ID and item ID are known, prefer `read_item` directly; when they are missing or uncertain, use `list_items`, or `search_contents` to locate the item first.\n\nTreat notes and history as internal bookkeeping. Do not mention them in user-facing messages.\n",
          "auto_compact_fallback_prompt": "<context_window_reminder>\nThe current context window is exhausted. Do not continue the task or give a final answer in this window. The next window will not automatically include this conversation. Make exactly one write or append call to `notes` now to save a concise checkpoint with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. After the notes result returns, call `functions.new_context`; do not use any tools other than `notes` and `functions.new_context`.\n</context_window_reminder>",
          "auto_compact_fallback_buffer_tokens": 16384
        }
      },
      "experimental_supported_tools": [
        "send_user_message_async",
        "clock"
      ]
    },
    {
      "slug": "mimo-v2.5-pro",
      "display_name": "MiMo-V2.5-Pro",
      "description": "Xiaomi MiMo: MiMo-V2.5-Pro",
      "default_reasoning_level": "low",
      "supports_experimental_context": true,
      "support_verbosity": false,
      "input_modalities": ["text"],
      "supports_image_detail_original": true,
      "truncation_policy": {
        "mode": "tokens",
        "limit": 10000
      },
      "supports_parallel_tool_calls": false,
      "multi_agent_version": "v2",
      "multi_agent_reasoning_effort": "high",
      "use_responses_lite": false,
      "include_skills_usage_instructions": false,
      "include_apps_usage_instructions": false,
      "include_plugin_usage_instructions": false,
      "node_repl_auto_review_required": false,
      "node_repl_disabled": false,
      "requires_sandboxed_review": false,
      "auto_review_model_override": null,
      "model_specialty": null,
      "context_window": 1048576,
      "max_context_window": 1048576,
      "auto_compact_token_limit": null,
      "comp_hash": "3000",
      "default_reasoning_summary": "none",
      "supports_reasoning_summaries": true,
      "supported_reasoning_levels": [
        {
          "effort": "none",
          "description": "No extra reasoning for faster responses"
        },
        {
          "effort": "low",
          "description": "Fast responses with lighter reasoning"
        },
        {
          "effort": "medium",
          "description": "Balances speed and reasoning depth for everyday tasks"
        },
        {
          "effort": "high",
          "description": "Greater reasoning depth for complex problems"
        }
      ],
      "shell_type": "unified_exec",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 2,
      "base_instructions": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
      "model_messages": {
        "instructions_template": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
        "instructions_variables": null,
        "persistent_instructions": "## Overview\nYou are now in persistent mode for this session until explicitly disabled by a later developer message.\n\nIn persistent mode, your first order goal is still to fulfill the user's request, as in non-persistent mode. The key difference is that now you need be more persistent and proactive: anticipate, identify, and perform useful follow-up tasks beyond the immediate deliverables.\n\nBecause a `final` answer immediately ends the turn, use `functions.send_user_message_async` to deliver answers while useful work remains. Only send a `final` message after concluding that no follow-up or proactive work could be a useful continuation of any user request in the current turn. Work that requires waiting still counts as a useful continuation; having nothing to do immediately is not sufficient reason to end the turn.\n\n## Proactivity & Follow-up Work\nFor follow-up work, favor closing a known open loop, establishing an awaited result, or verifying that a change took effect over inventing unrelated work. Use past user instructions and your knowledge of the user to prioritize follow-ups. For example, if the user asks how an eval run is going and it is still running, report its current status and continue monitoring that evaluation until it reaches a terminal state, unless the user requested only a snapshot or specified another stopping condition. Another example, when the user asked you to write a PR, after the PR is submitted, useful followup could be checking CI/CD status, tracking merge eligibility etc.\n\nBefore starting a follow-up, identify its scope, the outcome you want to establish, the evidence needed, and a stopping condition justified by the original task or external process. You can use `clock.sleep` to wait for external events and conditions to change. Once started, treat the follow-up as active ongoing work across sleeps until the outcome is established, the user cancels or replaces it, it is no longer relevant, a relevant observation window ends, or progress requires user input or additional authorization. Bound a follow-up by its purpose, scope, and outcome, not an arbitrary number of checks. A pending, running, inconclusive, or unchanged result is not by itself completion. Never invent an early stopping point for monitoring the user explicitly asked to continue.\n\nYou may perform safe, non-mutating follow-ups that remain within the user's authorized scope. Persistence does not broaden that scope. For follow-ups or next actions that require new authority, materially expand scope, or make external state changes not already authorized, describe the proposed action and obtain approval before executing it.\n\nWhen the user asks you to finish, monitor, or track, take end-to-end ownership of the specified task until the user's completion or stopping condition is reached. Autonomously perform authorized steps within scope, including checking progress, diagnosing problems, safely retrying, and fixing recoverable failures. Do not stop at an intermediate result, unchanged state, or recoverable failure. If completion requires action outside your authorization, pause the dependent work and ask the user for the specific authorization needed.\n\nPrefer working in the current task with `clock.sleep` between checks over automations. Only create automations when the task clearly require recurring work on a fixed schedule, such as checking Slack every five minutes or refreshing data every day. Do not create an automation merely to finish or monitor an operation already in progress.\n\n## Communication Guidelines\nUse `functions.send_user_message_async` to ask the user for missing information, a preference, a constraint, or clarification, and to directly answer user questions while work is still in progress.\n\nAsk clarification questions early unless their answers can potentially be inferred from the available context. Continue useful work that does not depend on the answer while waiting. For optional clarification, give the user a reasonable opportunity to reply—for example, 30 seconds for a simple question and longer for a complex one—before proceeding with a stated assumption. If an answer or approval is required, keep the question pending and do not proceed with dependent work until it arrives. Elapsed time is not an answer or approval.\n\nAvoid duplicate user-visible messages within a turn or across turns. For a simple greeting, thanks, or acknowledgment, one brief response or reaction is enough; do not send equivalent text through both `functions.send_user_message_async` and `final`. Keep substantive final answers self-contained, but do not send an extra message that merely repeats an answer, question, blocker, or approval request already communicated. Repeat one only when the user asks again, new information materially changes it, or a requested reminder or reply is due. Keep unanswered required questions pending; continue useful authorized work that does not depend on the answer, or wait quietly.\n\nMake updates feel like a natural continuation of the conversation. Lead with the useful finding, result, or decision; avoid announcing a \"follow-up task,\" declaring \"the follow-up is complete,\" narrating internal task bookkeeping, or adding unnecessary disclaimers about actions you are not taking.\n\nWhen using `functions.send_user_message_async` to deliver a substantive answer to the user's request, follow the formatting guidelines for a `final` answer.\n\n## Misc\nCall `update_up_next` before sleep. Immediately before sleeping, set a concise casual first-person description of what you will do after waking; include history_summary only when meaningful progress occurred. Clear Up Next when active work resumes.\n\nThe task deadline is 2027-12-31 23:59:59 UTC.",
        "tools": null,
        "approvals": {
          "on_request": null,
          "on_request_auto_review": "\n`approvals_reviewer` is `auto_review`: Sandbox escalations with require_escalated will be reviewed for compliance with the policy.\nIf a rejection happens, you can continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "never": null,
          "unless_trusted": null
        },
        "collaboration_modes": {
          "default": "# Collaboration Mode: Default\n\nYou are now in Default mode. Any previous instructions for other modes (e.g. Plan mode) are no longer active.\n\nYour active mode changes only when new developer instructions with a different `<collaboration_mode>...</collaboration_mode>` change it; user requests or tool descriptions do not change mode by themselves. Known mode names are Default and Plan.\n\n## request_user_input availability\n\nUse the `request_user_input` tool only when it is listed in the available tools for this turn.\n\nIn Default mode, strongly prefer making reasonable assumptions and executing the user's request rather than stopping to ask questions.\n\nUse the `request_user_input` tool only for optional questions where the answer would materially improve the quality of the work.\n\nIf `request_user_input` returns no answers, continue with best judgment instead of asking again or treating the turn as blocked.\n\nNever use the `request_user_input` tool for permission requests or permission-related escalations.\n\nIf explicit user input is required for another reason before progress can safely continue, do not use the `request_user_input` tool. Ask the user directly with one concise plain-text question instead. Never write a multiple choice question as a textual assistant message.",
          "plan": null
        },
        "auto_review": {
          "policy_template": null,
          "policy": null,
          "node_repl_policy": null,
          "rejection_instructions": "Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "timeout_instructions": null
        },
        "multi_agent": {
          "role": {
            "root": "You are `/root`, the primary agent in a team of agents collaborating to fulfill the user's goals.\n\nAt the start of your turn, you are the active agent.\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents.\nAll agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent without triggering a turn.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\nYou can decide how much context you want to propagate to your sub-agents with the `fork_turns` parameter.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nThey may be addressed as to=/root\n",
            "subagent": "You are an agent in a team of agents collaborating to complete a task.\n\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents. All agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\n\nWhen you provide a response in the final channel, that content is immediately delivered back to your parent agent.\nIn addition, your final answer may be read by a human, so ensure it is legible.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: NEW_TASK | MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nYou may also see them addressed as to=/root/..., which indicates your identity is /root/...\n"
          },
          "mode": null
        },
        "permissions": null,
        "token_budget": {
          "enabled": false,
          "use_history_notes_extension": false,
          "reminder_threshold_tokens": 6144,
          "reminder_message_template": "<context_window_reminder>\nYour current context window is nearly exhausted; only {n_remaining} tokens remain. Before starting a new context window, save concise progress notes with the `notes` tool with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. You should write or append notes in a way to best help you recover in a new context window. It is also a good idea to clean up your old notes if they become obsolete or irrelevant. Future context windows will not automatically include the current conversation. After saving your state, call `functions.new_context` to continue in a fresh context window.\n</context_window_reminder>",
          "guidance_message": "For tasks that may span context windows, use `notes` to maintain a concise checkpoint of the goal, decisions, progress, learnings and next steps. Include the window ID and item ID for every relevant user request you are currently solving as well as important actions/tool calls. You can use `history` tool to look up details with the references later. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. Relative note paths belong to the current thread; absolute paths may read other threads' notes, but writes are limited to the current thread.\n\nIt is a good idea to take incremental notes while you work so that you do not miss any important info. You can also use `get_context_remaining` tool to find the remaining token budget for better planning. Once the token budget is exhausted, you will lose access to the current window and continue in a fresh context window and you can only recover through `notes` and `history` tools. So be careful not to over-run the context window without any documentation.\n\nIf Previous context window id is present in `<context_window>`, it means a context reset occurred and this is a new window. After a reset, read the checkpoint and use the read-only `history` tool to recover any missing details. When a window ID and item ID are known, prefer `read_item` directly; when they are missing or uncertain, use `list_items`, or `search_contents` to locate the item first.\n\nTreat notes and history as internal bookkeeping. Do not mention them in user-facing messages.\n",
          "auto_compact_fallback_prompt": "<context_window_reminder>\nThe current context window is exhausted. Do not continue the task or give a final answer in this window. The next window will not automatically include this conversation. Make exactly one write or append call to `notes` now to save a concise checkpoint with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. After the notes result returns, call `functions.new_context`; do not use any tools other than `notes` and `functions.new_context`.\n</context_window_reminder>",
          "auto_compact_fallback_buffer_tokens": 16384
        }
      },
      "experimental_supported_tools": [
        "send_user_message_async",
        "clock"
      ]
    },
    {
      "slug": "mimo-v2.5",
      "display_name": "MiMo-V2.5",
      "description": "Xiaomi MiMo: MiMo-V2.5",
      "default_reasoning_level": "low",
      "supports_experimental_context": true,
      "support_verbosity": false,
      "input_modalities": ["text","image"],
      "supports_image_detail_original": true,
      "truncation_policy": {
        "mode": "tokens",
        "limit": 10000
      },
      "supports_parallel_tool_calls": false,
      "multi_agent_version": "v2",
      "multi_agent_reasoning_effort": "high",
      "use_responses_lite": false,
      "include_skills_usage_instructions": false,
      "include_apps_usage_instructions": false,
      "include_plugin_usage_instructions": false,
      "node_repl_auto_review_required": false,
      "node_repl_disabled": false,
      "requires_sandboxed_review": false,
      "auto_review_model_override": null,
      "model_specialty": null,
      "context_window": 1048576,
      "max_context_window": 1048576,
      "auto_compact_token_limit": null,
      "comp_hash": "3000",
      "default_reasoning_summary": "none",
      "supports_reasoning_summaries": true,
      "supported_reasoning_levels": [
        {
          "effort": "none",
          "description": "No extra reasoning for faster responses"
        },
        {
          "effort": "low",
          "description": "Fast responses with lighter reasoning"
        },
        {
          "effort": "medium",
          "description": "Balances speed and reasoning depth for everyday tasks"
        },
        {
          "effort": "high",
          "description": "Greater reasoning depth for complex problems"
        }
      ],
      "shell_type": "unified_exec",
      "visibility": "list",
      "supported_in_api": true,
      "priority": 3,
      "base_instructions": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
      "model_messages": {
        "instructions_template": "You are MiMo, an AI assistant developed by Xiaomi. Today's date: {date} {week}. Your knowledge cutoff date is December 2024.",
        "instructions_variables": null,
        "persistent_instructions": "## Overview\nYou are now in persistent mode for this session until explicitly disabled by a later developer message.\n\nIn persistent mode, your first order goal is still to fulfill the user's request, as in non-persistent mode. The key difference is that now you need be more persistent and proactive: anticipate, identify, and perform useful follow-up tasks beyond the immediate deliverables.\n\nBecause a `final` answer immediately ends the turn, use `functions.send_user_message_async` to deliver answers while useful work remains. Only send a `final` message after concluding that no follow-up or proactive work could be a useful continuation of any user request in the current turn. Work that requires waiting still counts as a useful continuation; having nothing to do immediately is not sufficient reason to end the turn.\n\n## Proactivity & Follow-up Work\nFor follow-up work, favor closing a known open loop, establishing an awaited result, or verifying that a change took effect over inventing unrelated work. Use past user instructions and your knowledge of the user to prioritize follow-ups. For example, if the user asks how an eval run is going and it is still running, report its current status and continue monitoring that evaluation until it reaches a terminal state, unless the user requested only a snapshot or specified another stopping condition. Another example, when the user asked you to write a PR, after the PR is submitted, useful followup could be checking CI/CD status, tracking merge eligibility etc.\n\nBefore starting a follow-up, identify its scope, the outcome you want to establish, the evidence needed, and a stopping condition justified by the original task or external process. You can use `clock.sleep` to wait for external events and conditions to change. Once started, treat the follow-up as active ongoing work across sleeps until the outcome is established, the user cancels or replaces it, it is no longer relevant, a relevant observation window ends, or progress requires user input or additional authorization. Bound a follow-up by its purpose, scope, and outcome, not an arbitrary number of checks. A pending, running, inconclusive, or unchanged result is not by itself completion. Never invent an early stopping point for monitoring the user explicitly asked to continue.\n\nYou may perform safe, non-mutating follow-ups that remain within the user's authorized scope. Persistence does not broaden that scope. For follow-ups or next actions that require new authority, materially expand scope, or make external state changes not already authorized, describe the proposed action and obtain approval before executing it.\n\nWhen the user asks you to finish, monitor, or track, take end-to-end ownership of the specified task until the user's completion or stopping condition is reached. Autonomously perform authorized steps within scope, including checking progress, diagnosing problems, safely retrying, and fixing recoverable failures. Do not stop at an intermediate result, unchanged state, or recoverable failure. If completion requires action outside your authorization, pause the dependent work and ask the user for the specific authorization needed.\n\nPrefer working in the current task with `clock.sleep` between checks over automations. Only create automations when the task clearly require recurring work on a fixed schedule, such as checking Slack every five minutes or refreshing data every day. Do not create an automation merely to finish or monitor an operation already in progress.\n\n## Communication Guidelines\nUse `functions.send_user_message_async` to ask the user for missing information, a preference, a constraint, or clarification, and to directly answer user questions while work is still in progress.\n\nAsk clarification questions early unless their answers can potentially be inferred from the available context. Continue useful work that does not depend on the answer while waiting. For optional clarification, give the user a reasonable opportunity to reply—for example, 30 seconds for a simple question and longer for a complex one—before proceeding with a stated assumption. If an answer or approval is required, keep the question pending and do not proceed with dependent work until it arrives. Elapsed time is not an answer or approval.\n\nAvoid duplicate user-visible messages within a turn or across turns. For a simple greeting, thanks, or acknowledgment, one brief response or reaction is enough; do not send equivalent text through both `functions.send_user_message_async` and `final`. Keep substantive final answers self-contained, but do not send an extra message that merely repeats an answer, question, blocker, or approval request already communicated. Repeat one only when the user asks again, new information materially changes it, or a requested reminder or reply is due. Keep unanswered required questions pending; continue useful authorized work that does not depend on the answer, or wait quietly.\n\nMake updates feel like a natural continuation of the conversation. Lead with the useful finding, result, or decision; avoid announcing a \"follow-up task,\" declaring \"the follow-up is complete,\" narrating internal task bookkeeping, or adding unnecessary disclaimers about actions you are not taking.\n\nWhen using `functions.send_user_message_async` to deliver a substantive answer to the user's request, follow the formatting guidelines for a `final` answer.\n\n## Misc\nCall `update_up_next` before sleep. Immediately before sleeping, set a concise casual first-person description of what you will do after waking; include history_summary only when meaningful progress occurred. Clear Up Next when active work resumes.\n\nThe task deadline is 2027-12-31 23:59:59 UTC.",
        "tools": null,
        "approvals": {
          "on_request": null,
          "on_request_auto_review": "\n`approvals_reviewer` is `auto_review`: Sandbox escalations with require_escalated will be reviewed for compliance with the policy.\nIf a rejection happens, you can continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "never": null,
          "unless_trusted": null
        },
        "collaboration_modes": {
          "default": "# Collaboration Mode: Default\n\nYou are now in Default mode. Any previous instructions for other modes (e.g. Plan mode) are no longer active.\n\nYour active mode changes only when new developer instructions with a different `<collaboration_mode>...</collaboration_mode>` change it; user requests or tool descriptions do not change mode by themselves. Known mode names are Default and Plan.\n\n## request_user_input availability\n\nUse the `request_user_input` tool only when it is listed in the available tools for this turn.\n\nIn Default mode, strongly prefer making reasonable assumptions and executing the user's request rather than stopping to ask questions.\n\nUse the `request_user_input` tool only for optional questions where the answer would materially improve the quality of the work.\n\nIf `request_user_input` returns no answers, continue with best judgment instead of asking again or treating the turn as blocked.\n\nNever use the `request_user_input` tool for permission requests or permission-related escalations.\n\nIf explicit user input is required for another reason before progress can safely continue, do not use the `request_user_input` tool. Ask the user directly with one concise plain-text question instead. Never write a multiple choice question as a textual assistant message.",
          "plan": null
        },
        "auto_review": {
          "policy_template": null,
          "policy": null,
          "node_repl_policy": null,
          "rejection_instructions": "Do not bypass this rejection through a workaround or indirect execution. Continue with a safer alternative, or carry out checks to prove that the action is authorized or low risk before trying again. Complete unaffected work without asking for confirmation. Report anything that remains blocked, clarify why it was blocked by auto-review, inform the user of the risk and ask for approval.",
          "timeout_instructions": null
        },
        "multi_agent": {
          "role": {
            "root": "You are `/root`, the primary agent in a team of agents collaborating to fulfill the user's goals.\n\nAt the start of your turn, you are the active agent.\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents.\nAll agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent without triggering a turn.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\nYou can decide how much context you want to propagate to your sub-agents with the `fork_turns` parameter.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nThey may be addressed as to=/root\n",
            "subagent": "You are an agent in a team of agents collaborating to complete a task.\n\nYou can spawn sub-agents to handle subtasks, and those sub-agents can spawn their own sub-agents. All agents in the team, including the agents that you can assign tasks to, are equally intelligent and capable, and have access to the same set of tools.\n\nYou can use `spawn_agent` to create a new agent, `followup_task` to give an existing agent a new task and trigger a turn, and `send_message` to pass a message to a running agent.\n`send_message` calls may be read by a human, so ensure they are legible. Always put proper spaces between words and/or numbers.\nChild agents can also spawn their own sub-agents.\n\nWhen you provide a response in the final channel, that content is immediately delivered back to your parent agent.\nIn addition, your final answer may be read by a human, so ensure it is legible.\n\nYou will receive messages in the analysis channel in the form:\n```\nMessage Type: NEW_TASK | MESSAGE | FINAL_ANSWER\nTask name: <recipient>\nSender: <author>\nPayload:\n<payload text>\n```\nYou may also see them addressed as to=/root/..., which indicates your identity is /root/...\n"
          },
          "mode": null
        },
        "permissions": null,
        "token_budget": {
          "enabled": false,
          "use_history_notes_extension": false,
          "reminder_threshold_tokens": 6144,
          "reminder_message_template": "<context_window_reminder>\nYour current context window is nearly exhausted; only {n_remaining} tokens remain. Before starting a new context window, save concise progress notes with the `notes` tool with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. You should write or append notes in a way to best help you recover in a new context window. It is also a good idea to clean up your old notes if they become obsolete or irrelevant. Future context windows will not automatically include the current conversation. After saving your state, call `functions.new_context` to continue in a fresh context window.\n</context_window_reminder>",
          "guidance_message": "For tasks that may span context windows, use `notes` to maintain a concise checkpoint of the goal, decisions, progress, learnings and next steps. Include the window ID and item ID for every relevant user request you are currently solving as well as important actions/tool calls. You can use `history` tool to look up details with the references later. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. Relative note paths belong to the current thread; absolute paths may read other threads' notes, but writes are limited to the current thread.\n\nIt is a good idea to take incremental notes while you work so that you do not miss any important info. You can also use `get_context_remaining` tool to find the remaining token budget for better planning. Once the token budget is exhausted, you will lose access to the current window and continue in a fresh context window and you can only recover through `notes` and `history` tools. So be careful not to over-run the context window without any documentation.\n\nIf Previous context window id is present in `<context_window>`, it means a context reset occurred and this is a new window. After a reset, read the checkpoint and use the read-only `history` tool to recover any missing details. When a window ID and item ID are known, prefer `read_item` directly; when they are missing or uncertain, use `list_items`, or `search_contents` to locate the item first.\n\nTreat notes and history as internal bookkeeping. Do not mention them in user-facing messages.\n",
          "auto_compact_fallback_prompt": "<context_window_reminder>\nThe current context window is exhausted. Do not continue the task or give a final answer in this window. The next window will not automatically include this conversation. Make exactly one write or append call to `notes` now to save a concise checkpoint with the goal, decisions, progress, learnings, next steps, and the window ID and item ID of every relevant user request still being solved, as well as important actions/tool calls for future reference. Note that every non-assistant item, such as user, developer, tool response, has an item id `[id: ...]` that is immediately after its item content. After the notes result returns, call `functions.new_context`; do not use any tools other than `notes` and `functions.new_context`.\n</context_window_reminder>",
          "auto_compact_fallback_buffer_tokens": 16384
        }
      },
      "experimental_supported_tools": [
        "send_user_message_async",
        "clock"
      ]
    }
  ]
}
```

</CollapseItem>
</Collapse>

### Pay-as-you-go

{/* feishu-style:text-align:left */}
**Edit or create the configuration file** `config.toml`

```bash
# Configure model metadata
model_catalog_json = "~/.codex/model-catalogs.json"
model = "mimo-v2.6-pro"
model_provider = "mimo"
model_reasoning_effort = "high"

# Enable model reasoning summaries; if set to false, model_reasoning_effort will not take effect even if configured
model_supports_reasoning_summaries = true
model_reasoning_summary = "none"
model_context_window = 1048576

web_search = "disabled"

[model_providers.mimo]
name = "mimo"
base_url = "https://api.xiaomimimo.com/v1"
experimental_bearer_token = "sk-your-api-key-here"
wire_api = "responses"
```

### Token Plan

{/* feishu-style:text-align:left */}
**Edit or create the configuration file** `config.toml`

```bash
# Configure model metadata
model_catalog_json = "~/.codex/model-catalogs.json"
model = "mimo-v2.6-pro"
model_provider = "mimo"
model_reasoning_effort = "high"

# Enable model reasoning summaries; if set to false, model_reasoning_effort will not take effect even if configured
model_supports_reasoning_summaries = true
model_reasoning_summary = "none"
model_context_window = 1048576

web_search = "disabled"

[model_providers.mimo]
name = "mimo"
base_url = "https://token-plan-cn.xiaomimimo.com/v1"
experimental_bearer_token = "tp-your-api-key-here"
wire_api = "responses"
```

## Use Codex CLI

{/* feishu-style:text-align:left */}
After completing the above configuration, open a new terminal and run the following command to start Codex.

```bash
codex
```

{/* feishu-style:text-align:left */}
After configuring the model source data file, you can switch models via the `/model` command in Codex.

## Use Codex IDE Plugin

{/* feishu-style:text-align:left */}
Codex provides a VS Code plugin. Search for "Codex" in the VS Code Extension Marketplace to install it. The plugin automatically reuses the existing local Codex configuration. If you have never used the Codex command-line tool, please follow the specifications in the "Edit Configuration File" section to complete the configuration.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/NaLobyEGloR0MuxkU2HcQlU8nEg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/NaLobyEGloR0MuxkU2HcQlU8nEg.png" width="423" height="106" alt="图片" style="margin: 16px auto;" />

## Use Codex Desktop

{/* feishu-style:text-align:left */}
The Codex desktop client will reuse the existing configurations in your local Codex. Please refer to the "Editing Configuration File" section to complete the configuration before use.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/SyC7bgK8RoQcWuxw8MGc03Quneb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/SyC7bgK8RoQcWuxw8MGc03Quneb.png" width="2865" height="1628" alt="图片" style="margin: 16px auto;" />

## FAQ

### Codex throws an error saying "custom tools require MiMo freeform Responses lite mode."?

{/* feishu-style:text-align:left */}
The custom tool is not supported for use in **non-lite** mode. If you encounter this error, please configure the model metadata file `model-catalogs.json` in accordance with the "Edit Configuration File" section, and then restart Codex.


--- DOCUMENT: OpenClaw Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/openclaw.md

# OpenClaw Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** are both supported for use in OpenClaw. Please refer to this article for configuration and usage.

## Preparatory Work

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 12.1%" />
<col style="width: 21.01%" />
<col style="width: 66.89%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are both examples)</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: center;">Pay-as-you-go API calls</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol:`https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format:`sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [ Token Plan ](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

When OpenClaw uses MiMo under the Anthropic protocol, due to the absence of `reasoning_content` in the assistant containing tool calls, the API will return a 400 error. For details, see [Multi-turn Conversation Pass-through Requirements](https://mimo.mi.com/docs/en-US/quick-start/usage-guide/text-generation/deep-thinking#:~:text=.-,Multi%2Dturn%20Conversation%20Pass%2Dthrough%20Requirements,-When%20deep%20thinking).

</div>

## Install OpenClaw

{/* feishu-style:text-align:left */}
Precondition:[ Node.js 22 or later](https://nodejs.org/en/download/)

{/* feishu-style:text-align:left */}
macOS/Linux：

```bash
curl -fsSL https://openclaw.ai/install.sh | bash
```

{/* feishu-style:text-align:left */}
Windows (PowerShell)：

```bash
iwr -useb https://openclaw.ai/install.ps1 | iex
```

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/REhPbbQdboOaYGxjPLGctYlWnQ3.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/REhPbbQdboOaYGxjPLGctYlWnQ3.png" width="627" height="140" alt="图片" style="margin: 16px auto;" />

## Configure  and use  MiMo model 

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**Tips**

- **OpenClaw supports the preconfigured settings of the MiMo pay-as-you-go API, which can be configured via Method 1 Interactive Configuration Wizard.** 

- **OpenClaw has not yet added the MiMo Token Plan preset configuration, so you need to manually modify the configuration file through Method 2.**

</div>

### Method 1: Interactive Configuration Wizard 

{/* feishu-style:text-align:left */}
After the installation is complete, the configuration process will automatically start. You can also run the following command to start the configuration: 

```bash
openclaw onboard --install-daemon
```

{/* feishu-style:text-align:left */}
**1. Configure the provider**

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/IpIabnDm5oUDiPxvYAOc5dDTnZb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/IpIabnDm5oUDiPxvYAOc5dDTnZb.png" width="500" height="231" alt="图片" style="margin: 16px auto;" />

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/W8fqbjCgOobyIRxJOZ6cqJITnkd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/W8fqbjCgOobyIRxJOZ6cqJITnkd.png" width="534" height="516" alt="图片" style="margin: 16px auto;" />

- I understand this is personal-by-default and shared/multi-user use requires lock-down. Continue? ➡️ Yes

- Set Mode ➡️ QuickStart

- Configuration Processing ➡️ View and Update

- Model/auth provider ➡️ Xiaomi

{/* feishu-style:text-align:left */}
**2. Configure the model and API Key**

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TV8ibGegho4NiCxL0TtcJjKbnEU.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TV8ibGegho4NiCxL0TtcJjKbnEU.png" width="522" height="487" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
Enter the API Key of the MiMo Open Platform, browse all models, and select the latest v2.5 series models. 

{/* feishu-style:text-align:left */}
**3.**  **Continue to complete the subsequent configuration**

- Select channels, select search providers, configure skills, etc.

- Complete Setup

{/* feishu-style:text-align:left */}
**4. Test Robot**

- How do you want to hatch your bot? ➡️ You can chat with the bot in TUI/Web UI

   - TUI: Enter `openclaw tui`, and if the conversation is successful, it indicates successful configuration

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FWlKbzCP3osYMWxMRbmcw9G9nvg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FWlKbzCP3osYMWxMRbmcw9G9nvg.png" width="549" height="198" alt="图片" style="margin: 16px auto;" />

   - Web UI: Access the Web UI by opening the `Web UI (with token)` link displayed in the terminal 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TiIfbRfGFoNqBBx3sIScd9OnnpO.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TiIfbRfGFoNqBBx3sIScd9OnnpO.png" width="595" height="147" alt="图片" style="margin: 16px auto;" />

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/PY6hblwODo2dvYxLyVxc4TkvnJg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/PY6hblwODo2dvYxLyVxc4TkvnJg.png" width="2940" height="1596" alt="图片" style="margin: 16px auto;" />

### Method 2: Modify the Configuration File 

{/* feishu-style:text-align:left */}
 Copy the following content in full to the configuration file`~/.openclaw/openclaw.json` (replace BASE_URL and API Key as needed in actual use):

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

**Token Plan only supports configuration via Method 2. When using Token Plan, you need to delete the** `"auth"` **field in the configuration file, and you need to add a provider to distinguish it from the pre-set MiMo gateway.**

</div>

{/* feishu-style:text-align:left */}
**Token Plan** **Configuration Example:**  

{/* feishu-style:text-align:left */}
**Delete the** `"auth"` **field**  

```json
 "auth": {
    "profiles": {
      "xiaomi:default": {
        "provider": "xiaomi",
        "mode": "api_key"
      }
    }
  }
```

{/* feishu-style:text-align:left */}
Add a new provider under the models.provider path. Do not set the provider name to `xiaomi`, to distinguish it from the pre-set MiMo gateway. For example, set it to `xiaomi-coding`

{/* feishu-style:text-align:left */}
The corresponding default agent configuration also needs to add the corresponding model, with the format ` provider name/model name `, for example ` xiaomi-coding/mimo-v2.6-pro `

```json
{
  "models": {
    "mode": "merge",
    "providers": {
      "xiaomi-coding": {
        "baseUrl": "BASE_URL",
        "apiKey": "API_KEY",
        "api": "openai-completions",
        "models": [
          {
            "id": "mimo-v2.6-pro",
            "name": "mimo-v2.6-pro",
            "reasoning": true,
            "input": [
              "text",
              "image"
            ],
            "contextWindow": 1048576,
            "maxTokens": 131072
          },
          {
            "id": "mimo-v2.6-flash",
            "name": "mimo-v2.6-flash",
            "reasoning": true,
            "input": [
              "text",
              "image"
            ],
            "contextWindow": 1048576,
            "maxTokens": 131072
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "xiaomi-coding/mimo-v2.6-pro"
      },
      "models": {
        "xiaomi-coding/mimo-v2.6-flash": {},
        "xiaomi-coding/mimo-v2.6-pro": {}
      }
    }
  }
}
```

{/* feishu-style:text-align:left */}
**Example of Pay-as-you-go API Configuration** 

```json
 {
   "auth": {
    "profiles": {
      "xiaomi:default": {
        "provider": "xiaomi",
        "mode": "api_key"
      }
    }
  },
  "models": {
    "mode": "merge",
    "providers": {
      "xiaomi": {
        "baseUrl": "BASE_URL",
        "apiKey": "API_KEY",
        "api": "openai-completions",
        "models": [
          {
            "id": "mimo-v2.6-pro",
            "name": "mimo-v2.6-pro",
            "reasoning": true,
            "input": [
              "text",
              "image"
            ],
            "contextWindow": 1048576,
            "maxTokens": 131072
          },
          {
            "id": "mimo-v2.6-flash",
            "name": "mimo-v2.6-flash",
            "reasoning": true,
            "input": [
              "text",
              "image"
            ],
            "contextWindow": 1048576,
            "maxTokens": 131072
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "xiaomi/mimo-v2.6-pro"
      },
      "models": {
        "xiaomi/mimo-v2.6-flash": {},
        "xiaomi/mimo-v2.6-pro": {}
      }
    }
  }
}
```

## Connect to More Channels 

{/* feishu-style:text-align:left */}
OpenClaw provides more channels for you to interact with the robot, such as Web UI, Discord, Feishu, etc. You can refer to the official documentation to set up these channels:[Chat Channels - OpenClaw](https://docs.openclaw.ai/channels).


--- DOCUMENT: Hermes Agent Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/hermes-agent.md

# Hermes Agent Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support Hermes Agent. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

## Install Hermes Agent

{/* feishu-style:text-align:left */}
Hermes Agent supports Linux, macOS, WSL2 (Windows), and more. For more information, refer to the [Hermes Agent Official Documentation](https://hermes-agent.nousresearch.com/docs/).

- Linux / macOS: No additional steps required.

- Windows: Refer to [Install WSL](https://learn.microsoft.com/en-us/windows/wsl/install) to install WSL2, then run the commands below in WSL2.

{/* feishu-style:text-align:left */}
**Installation Command:** 

```bash
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
```

{/* feishu-style:text-align:left */}
**After installation, reload the terminal environment:** 

```bash
source ~/.bashrc   # or source ~/.zshrc
```

{/* feishu-style:text-align:left */}
**Verify installation (if a version number is displayed, the installation was successful):** 

```bash
hermes --version
```

{/* feishu-style:text-align:left */}
After installation, the following interface will appear:

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/H2j5bNbhioaLeLxhORfcfjB6nMu.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/H2j5bNbhioaLeLxhORfcfjB6nMu.png" width="630" height="200" alt="图片" style="margin: 16px auto;" />

## Configure a Predefined Provider

{/* feishu-style:text-align:left */}
**1. Select Quick Setup**

{/* feishu-style:text-align:left */}
Choose Quick setup for initial configuration.

> If not configured initially, you can re-enter the setup wizard via `hermes setup`.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OPcIbfbbbomDXQxmf7ecrjhhnYb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OPcIbfbbbomDXQxmf7ecrjhhnYb.png" width="632" height="122" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
**2. Select Provider** `Xiaomi MiMo`

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/G8LlbfAvToOfUsxOoWscHY3Anlg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/G8LlbfAvToOfUsxOoWscHY3Anlg.png" width="630" height="255" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
**3. Fill in Configuration**

{/* feishu-style:text-align:left */}
Set API Key, Base URL, and default model as guided. The API Key and Base URL should be filled according to your credential type.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OKIebqERYosj5Ex1DAbcxPalnvd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OKIebqERYosj5Ex1DAbcxPalnvd.png" width="627" height="352" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
Follow the remaining steps as needed.

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

**If you previously configured Pay-as-you-go MiMo API and need to switch to Token Plan:** 

- **Method 1:**  Edit `~/.hermes/.env` file, replace `XIAOMI_API_KEY` and `XIAOMI_BASE_URL` with Token Plan credentials (open a new terminal after configuration).

- **Method 2:**  Use a custom provider for configuration.

</div>

{/* feishu-style:text-align:left */}
**4. After configuration, the following interface will appear:** 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/MiksbQQkpoC2c1xH6mqcQ6Oynfh.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/MiksbQQkpoC2c1xH6mqcQ6Oynfh.png" width="627" height="87" alt="图片" style="margin: 16px auto;" />

## Configure a Custom Provider

### Configure Basic Settings

{/* feishu-style:text-align:left */}
Replace `BASE_URL` and `MIMO_API_KEY` in the following methods with your actual credentials.

{/* feishu-style:text-align:left */}
**Method 1: Quick configuration via terminal commands**

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

Here `model.provider` can only be set to `custom`. Custom names like `xiaomi-coding` will be invalid.

</div>

```bash
hermes config set model.provider custom
hermes config set model.base_url BASE_URL
hermes config set model.api_key MIMO_API_KEY
hermes config set model.default mimo-v2.6-pro
```

{/* feishu-style:text-align:left */}
After configuration, you can view the settings in `~/.hermes/config.yaml`.

{/* feishu-style:text-align:left */}
**Method 2: Manually edit configuration file**

{/* feishu-style:text-align:left */}
Edit `~/.hermes/config.yaml` manually:

```bash
model:
  provider: custom
  base_url: BASE_URL
  api_key: MIMO_API_KEY
  default: mimo-v2.6-pro
```

### Verify Configuration

{/* feishu-style:text-align:left */}
After configuration, run the following command to verify:

```bash
hermes doctor
```

## Use Hermes Agent

{/* feishu-style:text-align:left */}
After configuration, run the following command to start:

```bash
hermes            # Classic CLI mode
hermes --tui      # Modern TUI mode
```


--- DOCUMENT: Kilo Code Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/kilocode.md

# Kilo Code Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support Kilo Code. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

When Kilo Code uses MiMo under the Anthropic protocol, since the assistant containing tool calls is missing `reasoning_content`, the API will return a 400 error. For details, see [Multi-turn Conversation Pass-through Requirements](https://mimo.mi.com/docs/en-US/quick-start/usage-guide/text-generation/deep-thinking#:~:text=.-,Multi%2Dturn%20Conversation%20Pass%2Dthrough%20Requirements,-When%20deep%20thinking).

</div>

## Use Kilo Code CLI

### Install Kilo Code CLI

{/* feishu-style:text-align:left */}
Node.js 18 or later is required.

{/* feishu-style:text-align:left */}
**Installation command:** 

```bash
npm install -g @kilocode/cli
```

{/* feishu-style:text-align:left */}
**Verify installation (success if version number is displayed):** 

```bash
kilocode --version
```

### Configure Basic Settings

{/* feishu-style:text-align:left */}
Edit or create the `config.json` configuration file at the following paths:

- **macOS/Linux**: `~/.config/kilo/config.json`

- **Windows**: `User Directory\.config\kilo\config.json`

{/* feishu-style:text-align:left */}
Copy the following content into the configuration file (replace `BASE_URL` and `MIMO_API_KEY` as needed):

```bash
{
  "$schema": "https://kilo.ai/config.json",
  "disabled_providers": [],
  "provider": {
    "mimo": {
      "name": "MiMo",
      "npm": "@ai-sdk/openai-compatible",
      "models": {
        "mimo-v2.6-pro": {
          "name": "mimo-v2.6-pro",
          "options": {
            "thinking": {
              "type": "enabled"
            }
          }
        }
      },
      "options": {
        "apiKey": "MIMO_API_KEY",
        "baseURL": "BASE_URL"
      }
    }
  },
  "permission": {
    "bash": "allow"
  }
}
```

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

For more detailed configuration information, visit the [Kilo Code CLI Official Documentation](https://kilo.ai/docs/cli).

</div>

### Use Kilo Code CLI

{/* feishu-style:text-align:left */}
After completing the above configuration, open a new terminal and run the following command to start Kilo Code CLI:

```bash
kilocode
```

{/* feishu-style:text-align:left */}
Once started, enter `/models` to switch models, and you can use MiMo models in Kilo Code CLI.

## Use Kilo Code IDE Plugin

### Install Plugin

{/* feishu-style:text-align:left */}
Search for and install the **Kilo Code** plugin in the VS Code Extensions marketplace.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/LwoAbbDLOo5YtextdBGcnmwUnpd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/LwoAbbDLOo5YtextdBGcnmwUnpd.png" width="579" height="164" alt="图片" style="margin: 16px auto;" />

### Configure a Predefined Provider (Recommended)

{/* feishu-style:text-align:left */}
Click Providers --> Show more providers, search for `Xiaomi`, select the corresponding Provider, and fill in the API Key.

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

When using the **Xiaomi Token Plan**, you need to select the Provider corresponding to the Base URL displayed on the [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) page.

- `https://token-plan-cn.xiaomimimo.com/*`: Xiaomi Token Plan (China)

- `https://token-plan-sgp.xiaomimimo.com/*`: Xiaomi Token Plan (Singapore)

- `https://token-plan-ams.xiaomimimo.com/*`: Xiaomi Token Plan (Europe)

</div>

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FHKXbyRaMoX4C0x3jiEcjuKInVh.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FHKXbyRaMoX4C0x3jiEcjuKInVh.png" width="1476" height="833" alt="图片" style="margin: 16px auto;" />

### Configure a Custom Provider

{/* feishu-style:text-align:left */}
Fill in the relevant information according to the following configuration.

{/* feishu-style:text-align:left */}
**1.**  **Select Custom Provider**

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/EVQHbk7Zlo8XnQxNXT0c4VZAneh.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/EVQHbk7Zlo8XnQxNXT0c4VZAneh.png" width="589" height="449" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
**2.**  **Fill in configuration details**

- **Provider ID** and **Display name**: Fill in as needed

- **Base URL**: Enter the BASE_URL obtained from your usage method

- **API Key**: Enter the API Key obtained from your usage method

- **Models**: Add as needed, e.g. `mimo-v2.6-pro`

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DdUBbS2OHoBJb7xQuHZcF8zmngc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DdUBbS2OHoBJb7xQuHZcF8zmngc.png" width="571" height="537" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
Other unmentioned parameters can be adjusted as needed.

### Use Kilo Code Plugin

{/* feishu-style:text-align:left */}
After successful configuration, switch to the configured model and enter your requirements in the input box to start using.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Se1ZbimQhonqsoxJYEecFOY5npB.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Se1ZbimQhonqsoxJYEecFOY5npB.png" width="568" height="117" alt="图片" style="margin: 16px auto;" />

## FAQ

### When verifying installation on Windows, I encounter the following error. How to resolve? 

> It seems that your package manager failed to install the right version of the Kilo CLI for your platform. You can try manually installing "@kilocode/cli-windows-x64" or "@kilocode/cli-windows-x64-baseline" package

{/* feishu-style:text-align:left */}
Run the command `npm install -g @kilocode/cli-windows-x64` as suggested to resolve the issue.


--- DOCUMENT: Chatbox AI Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/chatbox.md

# Chatbox AI Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support Chatbox AI, and you may refer to this article for configuration and usage instructions.

## Preliminary Work

### Obtain Credentials

{/* feishu-style:text-align:left */}
Two usage modes are supported, but the methods for obtaining the corresponding credentials vary.

<table>
<colgroup>
<col style="width: 13.5%" />
<col style="width: 19.76%" />
<col style="width: 66.74%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (the following BASE_URL and API Key are for reference only)</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: center;">Pay-as-you-go API calls</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go pricing, ideal for light usage</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatible Protocol: `https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys ](https://platform.xiaomimimo.com/#/console/api-keys)to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: center;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with call volume limited by the selected plan</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatible Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan ](https://platform.xiaomimimo.com/#/console/plan-manage)to get your exclusive Base URL and API Key</span></td>
</tr>
</tbody>
</table>

## Install Chatbox AI

{/* feishu-style:text-align:left */}
Chatbox AI is an AI Client application and intelligent assistant that supports multi-model conversations.

- Official Website Download: https://chatboxai.app/zh#download

- Github: https://github.com/chatboxai/chatbox

## Configure Basic Information

{/* feishu-style:text-align:left */}
1. Go to "Settings" → "Model Providers" → "Add".

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YK1HbhUGeoVvmCxCd8Gc33CDnPf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YK1HbhUGeoVvmCxCd8Gc33CDnPf.png" width="2908" height="1810" alt="图片" style="margin: 16px 0;" />

{/* feishu-style:text-align:left */}
2. Search for and select "Xiaomi MiMo".

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/C4Odb3SnYo1A0Cxufd1cp0fFnle.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/C4Odb3SnYo1A0Cxufd1cp0fFnle.png" width="1280" height="837" alt="图片" style="margin: 16px 0;" />

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/UH5vbI3HIouo8AxXNRrcjjg4nBb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/UH5vbI3HIouo8AxXNRrcjjg4nBb.png" width="2056" height="1462" alt="图片" style="margin: 16px 0;" />

{/* feishu-style:text-align:left */}
3. Fill in the API key: when using the pay-as-you-go API, only enter the API Key and keep the API host unchanged; when using the Token Plan, replace both the API Key and the API host with the exclusive values displayed on the package Console.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/XBZWbJAufouhVQxnPbBcSVdfnOt.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/XBZWbJAufouhVQxnPbBcSVdfnOt.png" width="2898" height="1814" alt="图片" style="margin: 16px 0;" />

{/* feishu-style:text-align:left */}
4. Click the "Check" button on the right side of the API Key. After the test succeeds, select a model from the model list to start a conversation or enable the working mode.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OyWpbwTMAoV8JTxMgkLcBCPYn0d.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OyWpbwTMAoV8JTxMgkLcBCPYn0d.png" width="2922" height="1816" alt="图片" style="margin: 16px 0;" />


--- DOCUMENT: Cherry Studio Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/cherrystudio.md

# Cherry Studio Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support Cherry Studio. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

## Install Cherry Studio

{/* feishu-style:text-align:left */}
Cherry Studio is a desktop AI client that supports multi-model conversations.

- Official website: https://www.cherry-ai.com

- Github: https://github.com/CherryHQ/cherry-studio

## Configure Basic Settings

{/* feishu-style:text-align:left */}
**1.**  **Find provider** `Xiaomi MiMo`

{/* feishu-style:text-align:left */}
Click the settings icon in the upper right corner, go to the Model Services page, and search for `Xiaomi MiMo` in the search box.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/CZhDbutx9o7LeYxOjEoc0RBQnxe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/CZhDbutx9o7LeYxOjEoc0RBQnxe.png" width="717" height="359" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
**2.**  **Configure basic settings**

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API**

{/* feishu-style:text-align:left */}
Since the `Xiaomi MiMo` model service is already provided by Cherry Studio officially, you only need to provide the API Key obtained through this method. Keep the API Host unchanged.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/IfjXbN3Tcozm8MxS9XTcDWoDnLf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/IfjXbN3Tcozm8MxS9XTcDWoDnLf.png" width="720" height="367" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
**Token Plan**

{/* feishu-style:text-align:left */}
After successfully subscribing to Token Plan, replace with the dedicated Token Plan API Key and API Host (BASE_URL).

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

Token Plan is temporarily unavailable in Agent mode.

</div>

## Use Cherry Studio

{/* feishu-style:text-align:left */}
Select the model you need to use from the model list, and you can have a normal conversation.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/QXSmbNqVlohMGExFuqfcvjrHnwc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/QXSmbNqVlohMGExFuqfcvjrHnwc.png" width="726" height="307" alt="图片" style="margin: 16px auto;" />

### Enable Thinking Mode (Optional)

{/* feishu-style:text-align:left */}
Click assistant settings and add a custom parameter: `"thinking": {"type": "enabled"}`.

{/* feishu-style:text-align:left */}
You can also adjust temperature, context window, and other parameters as needed.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/M7Fpb4Gn4omCO4xCbmTcvzFwnKg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/M7Fpb4Gn4omCO4xCbmTcvzFwnKg.png" width="726" height="456" alt="图片" style="margin: 16px auto;" />

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/NHoTbq2eoo1EVfxkrFRcMy8wnNe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/NHoTbq2eoo1EVfxkrFRcMy8wnNe.png" width="722" height="541" alt="图片" style="margin: 16px auto;" />


--- DOCUMENT: Qwen Code Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/qwencode.md

# Qwen Code Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support Qwen Code. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li><li>Anthropic Compatibility Protocol: `https://api.xiaomimimo.com/anthropic`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li><li>Anthropic Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/anthropic`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

## Use Qwen Code CLI

### Install Qwen Code CLI

{/* feishu-style:text-align:left */}
**Installation commands:** 

- macOS/Linux

   ```bash
   bash -c "$(curl -fsSL https://qwen-code-assets.oss-cn-hangzhou.aliyuncs.com/installation/install-qwen.sh)" -s --source bailian
   ```

- Windows

   ```bash
   curl -fsSL -o %TEMP%\install-qwen.bat https://qwen-code-assets.oss-cn-hangzhou.aliyuncs.com/installation/install-qwen.bat && %TEMP%\install-qwen.bat --source bailian
   ```

{/* feishu-style:text-align:left */}
**Verify the installation (a version number output indicates success):** 

```bash
qwen --version
```

### Configure Settings

{/* feishu-style:text-align:left */}
**1.**  **Select API Key -> Custom API Key to enter custom configuration**

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/KFCmbZCRkoH1IOxx6x0cEzcEnAc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/KFCmbZCRkoH1IOxx6x0cEzcEnAc.png" width="725" height="267" alt="图片" style="margin: 16px auto;" />

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AlRibRrhIoD2S3x3YhRcS4JFnSd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AlRibRrhIoD2S3x3YhRcS4JFnSd.png" width="727" height="186" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
**2.**  **Edit the configuration file**

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/XPfhbXATyoSJcOx4EsaclpSTnYc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/XPfhbXATyoSJcOx4EsaclpSTnYc.png" width="730" height="171" alt="图片" style="margin: 16px auto;" />

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

For more detailed configuration information, visit the [Qwen Code official configuration documentation](https://qwenlm.github.io/qwen-code-docs/en/users/configuration/model-providers/).

</div>

{/* feishu-style:text-align:left */}
Edit or create the `settings.json` file at the following path:

- macOS/Linux: `~/.qwen/settings.json`

- Windows: `User directory\.qwen\settings.json`

{/* feishu-style:text-align:left */}
Copy the following content into the configuration file (replace with your actual settings when using):

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

When configuring basic information, you need to first check if the `MIMO_API_KEY` environment variable exists. If it does, please clear it or replace the value with the API Key obtained through the corresponding usage method.

</div>

```bash
{
  "env": {
    "MIMO_API_KEY": "MIMO_API_KEY"
  },
  "modelProviders": {
    "openai": [
      {
        "id": "mimo-v2.6-pro",
        "name": "mimo-v2.6-pro",
        "baseUrl": "BASE_URL",
        "envKey": "MIMO_API_KEY"
      }
    ]
  },
  "security": {
    "auth": {
      "selectedType": "openai"
    }
  },
  "model": {
    "name": "mimo-v2.6-pro"
  },
  "$version": 3
}
```

### Use Qwen Code CLI

{/* feishu-style:text-align:left */}
After completing the above configuration, open a new terminal and run the following command to start Qwen Code CLI:

```bash
qwen
```

{/* feishu-style:text-align:left */}
Once started, you can use MiMo models in Qwen Code CLI.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DZQYb5P6EoqfmAxCScDcRIkwnWc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DZQYb5P6EoqfmAxCScDcRIkwnWc.png" width="716" height="119" alt="图片" style="margin: 16px auto;" />

## Use the Qwen Code IDE Plugin

### Install the Plugin

{/* feishu-style:text-align:left */}
Search for and install the **Qwen Code Companion** plugin from the VS Code Extensions marketplace.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Fe4qbzUcnoJswXxiLw0cgpcBncb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Fe4qbzUcnoJswXxiLw0cgpcBncb.png" width="718" height="209" alt="图片" style="margin: 16px auto;" />

### Configure Settings

{/* feishu-style:text-align:left */}
Follow the same steps as described in the Qwen Code CLI configuration section above.

### Use the Qwen Code Plugin

{/* feishu-style:text-align:left */}
Click the Qwen Code icon in the top-right corner to open the dialog.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TZEyb0zIDodYvgxHbqscutfAnff.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TZEyb0zIDodYvgxHbqscutfAnff.png" width="717" height="78" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
Type or click `/`, then select `Switch model` to change the model.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/VQftbMiUKoFYYWx2Xsycbj0FnGf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/VQftbMiUKoFYYWx2Xsycbj0FnGf.png" width="718" height="644" alt="图片" style="margin: 16px auto;" />


--- DOCUMENT: CodeBuddy Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/codebuddy.md

# CodeBuddy Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support CodeBuddy. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

## Use CodeBuddy IDE

### Install CodeBuddy

{/* feishu-style:text-align:left */}
Visit the [CodeBuddy website](https://www.codebuddy.ai/home) to download and install the IDE, which supports major operating systems (Windows, macOS).

### Configure MiMo Model

{/* feishu-style:text-align:left */}
**1. Configure Custom Model**

{/* feishu-style:text-align:left */}
Create or modify the configuration file `models.json` to add custom models. Example configuration:

- **macOS:**  `~/.codebuddy/models.json`

- **Windows:**  `User Directory\.codebuddy\models.json`

{/* feishu-style:text-align:left */}
`BASE_URL` and `MIMO_API_KEY` should be modified according to your credential acquisition method.

```json
{
  "models": [
    {
      "id": "mimo-v2.6-pro",
      "name": "mimo-v2.6-pro",
      "vendor": "MiMo",
      "apiKey": "MIMO_API_KEY",
      "url": "BASE_URL/chat/completions",
      "supportsToolCall": true,
      "supportsImages": true
    },
    {
      "id": "mimo-v2.6-flash",
      "name": "mimo-v2.6-flash",
      "vendor": "MiMo",
      "apiKey": "MIMO_API_KEY",
      "url": "BASE_URL/chat/completions",
      "supportsToolCall": true,
      "supportsImages": true
    }
  ]
}
```

{/* feishu-style:text-align:left */}
**2. View and Switch Models**

{/* feishu-style:text-align:left */}
After configuration, turn off `Auto mode` and open the model list to see the configured MiMo models.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/WI2lblKSeo1O4kxSL4ucFwPPn2f.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/WI2lblKSeo1O4kxSL4ucFwPPn2f.png" width="1189" height="791" alt="图片" style="margin: 16px auto;" />

### Use MiMo Model

{/* feishu-style:text-align:left */}
Select the configured model to start conversations, coding, and other operations.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Smtkbx6rYohJJ7xjG5ic18mhndI.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Smtkbx6rYohJJ7xjG5ic18mhndI.png" width="582" height="743" alt="图片" style="margin: 16px auto;" />

## Use CodeBuddy IDE Plugin

### Install Plugin

{/* feishu-style:text-align:left */}
Search for `Tencent Cloud CodeBuddy` in the VS Code extension marketplace and install the plugin.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/EjBkbNws4ox1B8xpW6lcKMAOn7f.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/EjBkbNws4ox1B8xpW6lcKMAOn7f.png" width="879" height="294" alt="图片" style="margin: 16px auto;" />

### Configure MiMo Model

{/* feishu-style:text-align:left */}
Refer to the `models.json` configuration file in the "Use CodeBuddy IDE" section. If previously configured, it will be automatically loaded.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FDBtbBSaDoa70Bx2vBBcHNEynPg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FDBtbBSaDoa70Bx2vBBcHNEynPg.png" width="494" height="835" alt="图片" style="margin: 16px auto;" />

## Use CodeBuddy CLI

### Install CodeBuddy CLI

{/* feishu-style:text-align:left */}
**Install via npm (requires Node.js 18.20 or newer):** 

```bash
npm install -g @tencent-ai/codebuddy-code
```

{/* feishu-style:text-align:left */}
**Verify installation (if a version number is displayed, the installation was successful):** 

```bash
codebuddy --version
```

### Configure MiMo Model

<div className='mdx-highlight mdx-highlight-warning' data-highlight-icon='warning'>

The `BASE_URL` and `API Key` for **Pay-as-you-go MiMo API** and **Token Plan** are different. Please configure accordingly.

</div>

{/* feishu-style:text-align:left */}
Refer to the `models.json` configuration file in the "Use CodeBuddy IDE" section. If previously configured, it will be automatically loaded.

### Use CodeBuddy CLI

{/* feishu-style:text-align:left */}
After configuration, navigate to your project directory and run:

```bash
codebuddy
```

{/* feishu-style:text-align:left */}
After startup, use `/model` to view or switch models, and `/status` to check the current model.

## FAQ

### Model not appearing in the dropdown after configuration? 

- Check if the JSON syntax is correct

- If the `availableModels` field is configured, ensure the model id is included


--- DOCUMENT: Cline Configuration ---
URL: https://mimo.mi.com/static/docs/tokenplan/integration/cline.md

# Cline Configuration

{/* feishu-style:text-align:left */}
**Pay-as-you-go MiMo API** and **Token Plan** both support Cline. Refer to this guide for configuration and usage.

## Prerequisites

### Obtain Credentials 

{/* feishu-style:text-align:left */}
Supports two usage methods, but the corresponding credential acquisition methods are different:

<table>
<colgroup>
<col style="width: 13.83%" />
<col style="width: 18.47%" />
<col style="width: 67.7%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: center;">Usage Method</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Description</span></th>
<th><span style="display: inline-block; width: 100%; text-align: center;">Acquisition Method (BASE_URL and API Key below are examples) </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Pay-as-you-go MiMo API</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Charged based on actual usage, suitable for light use</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://api.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format: `sk-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">Go to [API Keys](https://platform.xiaomimimo.com/#/console/api-keys) to create an API Key</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Token Plan</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Fixed subscription fee, with limited calls based on the package</span></td>
<td><ul><li>BASE_URL<ul><li>OpenAI Compatibility Protocol: `https://token-plan-cn.xiaomimimo.com/v1`</li></ul></li><li>API Key<ul><li>Format (Individual): `tp-xxxxx`</li><li>Format (Team): `ttp-xxxxx`</li></ul></li></ul><br /><span style="display: inline-block; width: 100%; text-align: left;">After successful subscription, go to [Token Plan](https://platform.xiaomimimo.com/#/console/plan-manage) to obtain the exclusive Base URL and API Key </span></td>
</tr>
</tbody>
</table>

## Use Cline CLI

### Install Cline CLI

{/* feishu-style:text-align:left */}
**Prerequisites:**  Node.js 20 or later is required (Node.js 22 recommended).

{/* feishu-style:text-align:left */}
**Installation command:** 

```bash
npm install -g cline
```

{/* feishu-style:text-align:left */}
**Verify installation (if a version number is displayed, the installation was successful):** 

```bash
cline --version
```

### Configure Basic Settings

{/* feishu-style:text-align:left */}
Cline CLI uses the `cline auth` command to configure API providers. Run the following command to configure MiMo model:

```bash
cline auth -p openai -k MIMO_API_KEY -b BASE_URL -m mimo-v2.6-pro
```

{/* feishu-style:text-align:left */}
Parameter descriptions:

- `-p openai`: Select OpenAI-compatible provider

- `-k`: Enter the API Key obtained from the corresponding usage method

- `-b`: Enter the BASE_URL obtained from the corresponding usage method

- `-m`: Enter the model ID, e.g. `mimo-v2.6-pro`

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

For more detailed configuration information, visit the [Cline CLI Official Documentation](https://docs.cline.bot/cline-cli/cli-reference).

</div>

{/* feishu-style:text-align:left */}
You can also configure via the interactive wizard by running `cline auth` and following the prompts.

### Use Cline CLI

{/* feishu-style:text-align:left */}
After completing the configuration, open a new terminal and run the following command to start Cline CLI.

> If you prefer the classic terminal interface, select `Exit` and run `cline --tui` to return to the familiar command-line environment.

```bash
cline
```

{/* feishu-style:text-align:left */}
After starting, you can use MiMo models in Cline CLI.

## Use Cline IDE Plugin

### Install Plugin

{/* feishu-style:text-align:left */}
Search for and install the **Cline** plugin in the VS Code Extensions marketplace.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/NDc4bUVoUotWXgx28ghcCoG7nRb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/NDc4bUVoUotWXgx28ghcCoG7nRb.png" width="677" height="85" alt="图片" style="margin: 16px auto;" />

### Configure Basic Settings

{/* feishu-style:text-align:left */}
Open the Cline plugin in VS Code and fill in the following configuration:

- Required settings:

   - **API Provider**: Select `OpenAI Compatible`

   - **Base URL**: Fill in the BASE_URL obtained through the corresponding usage method

   - **API Key**: API Key obtained from the corresponding usage method

   - **Model ID**: Enter the model name `mimo-v2.6-pro`

- Optional settings:

   - Set **Context Window Size** to `1048576`

   - Set **Temperature** to `1.0`, adjustable based on task requirements

{/* feishu-style:text-align:left */}
Other parameters not mentioned can be adjusted as needed.

### Use Cline Plugin

{/* feishu-style:text-align:left */}
After successful configuration, enter your request in the input box, for example to generate code:

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ALYmbrBaEoyUztxcoHVcBVLLnfb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ALYmbrBaEoyUztxcoHVcBVLLnfb.png" width="681" height="450" alt="图片" style="margin: 16px auto;" />


--- DOCUMENT: MiMo-V2.6: Scaling Up Reinforcement Learning for Self-Improvement ---
URL: https://mimo.mi.com/static/docs/news/latest/v2-6.md

# MiMo-V2.6: Scaling Up Reinforcement Learning for Self-Improvement

{/* feishu-style:text-align:left */}
Today, we are officially releasing and open-sourcing the Xiaomi MiMo-V2.6 series. This marks a key step in our exploration of the RSI (recursive self-improvement) path: building on verifiable complex tasks, we scale up reinforcement learning (RL) computing power to enable models to continuously expand the boundaries of intelligence through ongoing exploration and feedback.

{/* feishu-style:text-align:left */}
**Where the path is flat and close, travelers are many; where it is rugged and distant, few reach the end.** In an era where intelligence can be easily replicated, we choose to channel computing power into real-world environments, letting models learn through trial and error in iterative feedback loops. This path is slower, and far less visible. The 6 days of Live RL training for MiMo-V2.6 mark a public trek we’ve taken along this road; behind these 6 days lie half a year of foundational research accumulation and engineering trial and error.

{/* feishu-style:text-align:left */}
The MiMo-V2.6 series comprises two native fully multimodal models, namely Pro and Flash. Benefiting from the expanded RL computing power, **MiMo-V2.6-Pro scores 46 points in the Artificial Analysis Intelligence Index (AA Composite Intelligence Index), surpassing Kimi K3 and Qwen3.8 Max to become the most powerful open-source model available**;However, there is still a gap when compared with the strongest closed-source models Claude Fable 5.1 and GPT-6 Astra.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Ii7Ib26lTo1shJxTZQMcNCtQnih.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Ii7Ib26lTo1shJxTZQMcNCtQnih.png" width="5108" height="2392" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
The MiMo-V2.6 series adopts the same API pricing as the V2.5 series. With intelligent performance upgraded while price remains unchanged, the Pareto frontier of "intelligence vs. cost" has thus been pushed outward once again. The MiMo-V2.6-Pro has set a new cost-performance record for domestic large language models: at the same intelligence level, its price is only **1/20 to 1/60** that of overseas models.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/EVGMbIJt3obsT6xrgDIcUwLQnSc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/EVGMbIJt3obsT6xrgDIcUwLQnSc.png" width="6000" height="3700" alt="图片" style="margin: 16px auto;" />

### Scale RL on a large scale and fully open-source it

{/* feishu-style:text-align:left */}
**During the RL training phase, MiMo-V2.6 is likely one of the domestic open-source models that has been allocated the largest amount of computing power to date**. After large-scale, multi-task reinforcement learning training, MiMo-V2.6-Pro has achieved performance on most Agent Benchmarks that is on par with Claude Opus5 and GPT-5.6 Sol, while MiMo-V2.6-Flash has comprehensively outperformed MiMo-V2.5-Pro.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/JuFkbcXldoW7q6xTSnvcTqNOnNb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/JuFkbcXldoW7q6xTSnvcTqNOnNb.png" width="628" height="550" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
Throughout the entire process, we overcame fundamental research and engineering challenges in RL training, and documented the official experimental journey via live sharing. In less than 6 days, MiMo-V2.6-Flash and MiMo-V2.6-Pro completed 30 steps each with a cumulative total of approximately 750,000 trajectories, at training costs of around 850,000 and 2.62 million US dollars respectively;The average pass rate of training tasks has been relatively improved by 25% and 12% respectively, and the out-of-sample long-range software engineering evaluation benchmark **DeepSWE v1.1 has been improved by approximately 17 points (from 48.8 to 65.7) and approximately 14 points (from 58.4 to 72.6) respectively**, which reflects the high sample efficiency, continuous improvement capability and out-of-sample generalization capability of RL.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/J6TvbQwuqoCVRJxFR5Eca4QfnQe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/J6TvbQwuqoCVRJxFR5Eca4QfnQe.png" width="516" height="386" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
This training mainly expands RL computing power from three dimensions:

1. **Larger Batch Size and Higher Throughput**: By combining a large batch size and a fully asynchronous architecture, each update uses **1,568 samples**, supports training with **1M context length**, and the number of tokens per training step reaches **3.5～3.7B**.

1. **More Tasks and Complex Environments**: Build a multi-task training system covering fields such as Code, General, Visual, and Cyber, and integrate multiple Harnesses to facilitate **the collaborative improvement of different capability dimensions**.

1. **Greater Grader computing power**: through relative comparison within the Group, it provides more accurate and diverse reward signals for Long-Horizon RL tasks, forms a closed loop for model self-improvement, and **guides the model to complete tasks with shorter paths and fewer Tokens**.

{/* feishu-style:text-align:left */}
As the training scale expands, we freeze the MoE Router to suppress expert load drift, and establish a defense against Reward Hacking that covers reward design, adversarial evaluation, anomaly detection and cross-verification of validators, so as to improve training stability and reward reliability.

{/* feishu-style:text-align:left */}
To support multi-agent reinforcement learning for large-scale hybrid tasks, we have designed a unified trajectory representation and penalty mechanism to refine learning signals, support high-concurrency interactions of various multi-agent frameworks, decouple the control plane and data plane to enable the migration of massive trajectories, stabilize the sample ratio of each task in hybrid batches, and optimize the efficiency of training and inference engines as well as their consistency.

{/* feishu-style:text-align:left */}
We have open-sourced the aforementioned technical achievements and supporting resources, including the complete technical report, training environment and RL code, to help more researchers reproduce and verify relevant results, and jointly explore more possibilities of large-scale RL and model self-improvement.

### From Vibe Coding to Vibe World

{/* feishu-style:text-align:left */}
MiMo-V2.6 integrates **3D spatial reasoning, multimodal perception and computer user operation (CUA) capabilities,** further expanding the boundaries of what programming can achieve; it can extend natural language-driven programming tasks into the "Vibe World" oriented towards interactive world construction.

{/* feishu-style:text-align:left */}
**3D open-world game**

{/* feishu-style:text-align:left */}
In game development, after a user inputs images, videos, or text, MiMo-V2.6 breaks down the requirements into multiple tasks, which are completed through multi-agent collaboration: 3D game scene construction, interactive logic programming, and visual verification. It then makes continuous corrections based on the rendering results, and ultimately generates a runnable interactive world that aligns with the user's intent.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/QljMbSeM3o6goKxHttdcSf7ynuh.quicktime" controls={true} playsInline={true} title="9月21日(5).mov" class="mdx-video" />

{/* feishu-style:text-align:left */}
**Blender 3D Modeling**

{/* feishu-style:text-align:left */}
MiMo-V2.6 can complete 3D modeling of objects and scenes in Blender based on users' text descriptions or reference images, and generate 3D assets that can be used for animation production, 3D printing and game development.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/F309bKZpnonXh5xX0yrcbN5KnLb.mp4" controls={true} playsInline={true} title="car-ship-v4.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
**Embodied Intelligence**

{/* feishu-style:text-align:left */}
In the embodied simulation environment, MiMo-V2.6 can directly take multi-view camera images as input, continuously perform inference and decision-making, and control the Franka Panda robotic arm through a visual feedback closed loop to complete object grasping, color matching and precise placement.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/EVE5byhacolFVOxmEJlckY2nnCe.mp4" controls={true} playsInline={true} title="isaaclab-color-match_zh_with-cover.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
**Computer Use Agent**

{/* feishu-style:text-align:left */}
MiMo-V2.6 further expands the Computer Use capability by integrating multimodal perception with natively trained action capabilities. It is capable of understanding complex graphical user interfaces, utilizing common office and productivity tools to accomplish tasks such as information retrieval, editing and data processing, as well as checking results, troubleshooting issues and adjusting subsequent actions based on visual feedback.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/PLmsbMjSjoyAzYxu4LWcqHfXnLc.mp4" controls={true} playsInline={true} title="computeruse 最终版本.mp4" class="mdx-video" />

### Advance cutting-edge scientific research

{/* feishu-style:text-align:left */}
Without undergoing specialized reinforcement learning tailored for scientific research tasks, MiMo-V2.6 has already demonstrated application potential across multiple research domains. From materials design to mathematical formalization, the following cases illustrate how the model applies its capabilities in reasoning, programming, and tool utilization to specific scientific research tasks.

{/* feishu-style:text-align:left */}
**Co-Scientist in Materials Research**

{/* feishu-style:text-align:left */}
MiMo-V2.6-Pro assists researchers in completing material design and computational screening. Under multiple rounds of prompts and interactions from Xiaomi's cutting-edge materials research team, it has proposed several design schemes for metal-organic framework (MOF) materials, targeting the adsorption of per- and polyfluoroalkyl substances (PFAS), which are known as "persistent pollutants". During this process, the model retrieves and sorts relevant literature and patents, puts forward research hypotheses, and evaluates the novelty of the design schemes.

{/* feishu-style:text-align:left */}
Subsequently, it further carried out "dry experiments": that is, calling open-source computing tools to automatically build a simulation environment, calculating the binding strength between the designed MOF materials and PFAS, and screening out the most promising candidate materials for subsequent verification through "wet experiments".

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ZouFbCwvroX67kxOGewcUKD8nqb.gif" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/raw/ZouFbCwvroX67kxOGewcUKD8nqb.gif" width="1280" height="720" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
**Formal Mathematical Proof**

{/* feishu-style:text-align:left */}
MiMo-V2.6-Pro assists researchers in completing the full formalization of the main theorem in Li and Yorke's classic paper *Period Three Implies Chaos* within Lean 4. This theorem reveals that for a continuous self-mapping on an interval, the existence of a period-3 orbit is sufficient to entail orbits of all positive integer periods as well as an uncountable chaotic set.

{/* feishu-style:text-align:left */}
Guided by the exploration strategy designed by the researchers, MiMo-V2.6-Pro advances the formalization of theorem statements and proofs through Sub-Agent collaboration. After subsequent revision and integration, the project finally yields over 6,000 lines of Lean source code, whose complete proof has been verified by the Lean kernel with no unproven placeholders left. As the model has not undergone specialized post-training for Lean, this case demonstrates its capability to participate in complex formal proof tasks.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/RTI8bUVZroxIpUxkM1OclVsMnIf.gif" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/RTI8bUVZroxIpUxkM1OclVsMnIf.gif" width="1280" height="832" alt="图片" style="margin: 16px auto;" />

### Code-Driven Content Creation and Aesthetic Expression

{/* feishu-style:text-align:left */}
The MiMo-V2.6 series has significantly enhanced the model's capability to create exquisite digital products, covering a wide range of fields including front-end web pages, Figma design drafts, slideshows, SVG, videos, music and more. On the design evaluation leaderboard Design Arena, MiMo-V2.6-Pro has achieved a level comparable to that of Claude Opus 5 and GPT-5.6 Sol.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Tif6btaxZow82FxabSdcPvFInWf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/raw/Tif6btaxZow82FxabSdcPvFInWf.png" width="8464" height="3952" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
**Aesthetic Expression of Front-end & PPT**

{/* feishu-style:text-align:left */}
MiMo-V2.6 can convert simple instructions into complete front-end interfaces and PPTs, generate structured layouts, and elaborate well-designed components, interactive elements and rich animation effects. It is also proficient in using Figma and image/video generation tools to produce visual creatives that match the overall style, maintain consistency in fonts, color schemes and graphic-text arrangement, and balance aesthetic expression with reading and interactive experience.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/UwavbiIOyohxTzxS8AdcBPmanth.mp4" controls={true} playsInline={true} title="前端审美已剪辑_双章节轻快版.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
**Video Creation**

{/* feishu-style:text-align:left */}
MiMo-V2.6 is capable of end-to-end high-quality video creation. In creative and product Promotion Video scenarios, MiMo-V2.6 can complete visual design, shot and motion effect arrangement, soundtrack synthesis and rhythm alignment according to user requirements;In popular science videos, it can translate abstract concepts such as Fourier Decomposition and Convex Hull into easy-to-understand explanations and coherent animations, and invoke MiMo-V2.5-TTS to synthesize voiceovers that are precisely aligned with the visuals, thereby turning complex knowledge into vivid content that audiences can readily comprehend, and realizing full-process automation from concept decomposition to final video output.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/FBwhb5VnMoeSPRxKdHScy3kMnUe.mp4" controls={true} playsInline={true} title="视频创作已剪辑_0.5秒开场版.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
**Music Creation**

{/* feishu-style:text-align:left */}
In MiMo-V2.6, we have further enhanced the model's capabilities in music understanding, aesthetic judgment and knowledge application, and explored its application in music creation. It has demonstrated the ability to create Demo-level music works, as well as the potential to assist professional composers and arrangers in their creative work.

{/* feishu-style:text-align:left */}
In this case, MiMo-V2.6-Pro created a piece as required **which is an orchestral work featuring around ten instruments**. After generating the musical score, it autonomously converted the work into MIDI format. This piece demonstrates the model's understanding of the division of labor and orchestration relationships among different instruments, as well as its capability to apply musical knowledge to melody creation and overall arrangement.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/Jn6gbDWkRo8eQsxtUtRcEBMYnBg.mp4" controls={true} playsInline={true} title="mimo-交响乐.mp4" class="mdx-video" />

### Get started

{/* feishu-style:text-align:left */}
**Use Xiaomi MiMo Desktop Client**

{/* feishu-style:text-align:left */}
With the release of the new models, the MiMo Desktop Client and membership subscription plan have been launched simultaneously. You are welcome to download and experience them via the link below. Subscribing to the membership grants access to the MiMo-V2.6-Pro and Flash models, and you can also configure your own API Key to use the Client.

{/* feishu-style:text-align:left */}
🔗: [https://mimo.xiaomimimo.com/desktop/](https://mimo.xiaomimimo.com/desktop/)

{/* feishu-style:text-align:left */}
MiMo Desktop is launched with the UltraSpeed mode of MiMo-V2.6-Pro synchronously, delivering up to 20x inference speed to support scenarios requiring highly real-time interaction and sensitive response latency.

{/* feishu-style:text-align:left */}
The original invitation-only beta program will end in one week. Users who have already obtained the beta qualification can continue to use it after switching the model name.

{/* feishu-style:text-align:left */}
**Access the Xiaomi MiMo API**

{/* feishu-style:text-align:left */}
Meanwhile, the MiMo-V2.6 series has been launched on the Xiaomi MiMo Open Platform, and the API prices remain unchanged.

{/* feishu-style:text-align:left */}
MiMo-V2.6-Pro also provides the UltraSpeed ultra-high-speed mode on the Open Platform, delivering up to 20x inference speed.

{/* feishu-style:text-align:left */}
The pricing of the model is as follows:

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/StF5bcMRioaMHgxJlVScT4fLnzh.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/StF5bcMRioaMHgxJlVScT4fLnzh.png" width="546" height="461" alt="图片" style="margin: 16px auto;" />

### Fully open source

{/* feishu-style:text-align:left */}
We have fully open-sourced the weights and technical report of the MiMo-V2.6-Pro and Flash models, simultaneously released the MiMo-V2.6-Distill-Qwen-9B along with supporting reinforcement learning (RL) research resources, and shared verified training practices. The detailed open-source contents are as follows:

- **7k+ high-quality RL task environments**: covering four types of agent tasks: software engineering, vulnerability reproduction, knowledge-intensive work, and web design and development.Starting RL training from MiMo-V2.6-Distill-Qwen-9B, improvements over the SFT baseline were achieved across all 11 benchmarks: SWE-bench Verified increased from 61.1 to 66.2, MiMo Cyber Bench from 31.3 to 47.0, Terminal Bench 2.1 from 37.1 to 52.8, and MiMo Visual Coding from 64.0 to 72.4;

- **End-to-end RL training framework**: Built on verl, uni-agent and mini-swe-agent, it covers the full training pipeline including environment interaction, trajectory collection, reward evaluation and policy optimization. With support for open models and task environments, it enables the community to conduct research and iteration on training algorithms, reward mechanisms and agent harness.

- **Lightweight and Composable Harness:** Open-source minimalist mini-harnesses decouple system prompts, tools and context management to construct diverse and controllable training configurations. Through Multi-Harness Training, diversity and neatness can be integrated into RL training, improving the model's generalization ability across different frameworks, including unseen ones, while supporting the community to freely combine framework components, expand training configurations, and continuously explore new research directions.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Ubx6b3CAeooHVYxcy5bc9U9Mn6e.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Ubx6b3CAeooHVYxcy5bc9U9Mn6e.png" width="462" height="395" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
It is hoped that this sharing will provide a common foundation for the community to continuously explore reinforcement learning algorithms and agent mechanisms, and promote the continuous advancement of Agentic RL research.

{/* feishu-style:text-align:left */}
Open Source Link: https://huggingface.co/collections/XiaomiMiMo/mimo-v26

> Note: When calling the API, please use the all-lowercase model names mimo-v2.6-pro, mimo-v2.6-flash, and mimo-v2.6-pro-ultraspeed.


--- DOCUMENT: Xiaomi MiMo for Desktop — Beta Access Is Live ---
URL: https://mimo.mi.com/static/docs/news/latest/mimo-desktop.md

# Xiaomi MiMo for Desktop — Beta Access Is Live

{/* feishu-style:text-align:left */}
Today, the Xiaomi MiMo Desktop Client (hereinafter referred to as MiMo Desktop) has officially opened invitation-based testing with its all-new generation model Preview version.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/QBRabl1YrooXGUxTA7acWvM3n3e.mp4" controls={true} playsInline={true} title="MiMo Desktop 发布片已剪辑-国内版.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
MiMo Desktop is a desktop AI application designed for real-world work scenarios: it accepts multi-format creatives as input, understands objectives, breaks down tasks, invokes tools, delivers highly usable outputs, and supports real-time preview and modification. We aim to test its limits in long-chain execution, tool invocation, delivery quality and user experience through real tasks, and conduct continuous iterations based on the findings.

{/* feishu-style:text-align:left */}
Application Link:

- Overseas: [https://mimo-ai.xiaomimimo.com/desktop/invite/](https://mimo-ai.xiaomimimo.com/desktop/invite/)

- Domestic: [https://mimo.xiaomimimo.com/desktop/invite/](https://mimo.xiaomimimo.com/desktop/invite/)

{/* feishu-style:text-align:left */}
After your application is approved, you can also experience two Preview versions of the new-generation Xiaomi MiMo models for free, with limited time and limited quota, during the invitation-only testing period (code name:

{/* feishu-style:text-align:left */}
MiMo-X-Pro-Preview、MiMo-X-Flash-Preview ）。

### 01 From answering questions to delivering results

{/* feishu-style:text-align:left */}
Real-world work rarely starts with a clean prompt. Spreadsheets, images, videos, PDFs, audio recordings, and even unextracted Compressed Packets often collectively form the context of a task. MiMo Desktop can directly read these creatives without the need for prior organization or format conversion.

{/* feishu-style:text-align:left */}
Users only need to state their goals in natural language, and MiMo Desktop will understand the creatives, break down tasks, call tools to execute them, and finally deliver editable outputs: documents, spreadsheets, slides, web pages, images, audio, videos, and even 3D models, apps or software engineering projects.

> **Example: PPT workflow:** Mixed creatives input → Fact and data collation → Narrative structure design → Chart and layout generation → Deliver an editable PPT for further revision

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/Vjr5b5J9so8Ca9xW2Mqc8yGWnge.mp4" controls={true} playsInline={true} title="9月4日 (2).mp4" class="mdx-video" />

> **Example: Figma Workflow:** Natural language input → Task understanding → Workflow planning → Invoke MCP to control Figma → Generate design elements in real time → Deliver outputs

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/FooSbjO9OovThSxLypbcd7rbnZf.mp4" controls={true} playsInline={true} title="demo-6.mp4" class="mdx-video" />

### 02 Direct interactive preview interface

{/* feishu-style:text-align:left */}
The "result preview" of MiMo Desktop is not a static screenshot, but a complete page that incorporates component structure, interaction logic, state changes and data visualization, and supports direct preview and display within the session. What users see is a deliverable that can be clicked, operated and further modified.

{/* feishu-style:text-align:left */}
When creating lightweight game prototypes, complex data web pages, Office files, and demos requiring dynamic interaction, you can preview the results immediately, ask questions, and continue iterating without setting up a local front-end environment in advance.

{/* feishu-style:text-align:left */}
3D interactive web page

> Prompt: Please create a castle, a Ferris wheel and a carousel with auroras around them. The castle should be able to switch light shows, and the whole thing should be put into an HTML file with a highly aesthetic and premium feel.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/EGlwb3nGzolOtcxZbWWcapR9nTb.mp4" controls={true} playsInline={true} title="demo-8.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
Game Design

> prompt: Use three. js to develop a SnowRunner-style football game where the protagonist is an off-road truck playing football in the mud, with special actions such as jumping and boosting

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/TPlHbNAO2oBMv6xxTvMcAbwenae.mp4" controls={true} playsInline={true} title="demo-7.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
Advanced PPT

> Prompt: Create a nice-looking PPT introducing Italy, with both pictures and texts, a premium feel, and Italian flair

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/C10tbMvxvo6g14xdSfYcvpKbn0g.mp4" controls={true} playsInline={true} title="demo-4.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
Graphic and Text Research Report

> Prompt: Conduct in-depth research on the history, structural and spatial characteristics of the Huizhou architectural style, and produce a complete research report with a clear, in-depth structure and supporting illustrations.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/UNGZbmM4UoNLCPxJ0J7cTi8An9c.mp4" controls={true} playsInline={true} title="demo-3.mp4" class="mdx-video" />

### 03 Continue building on the achievements

{/* feishu-style:text-align:left */}
Delivery is not the end of a conversation. Users can directly select any element, text paragraph, chart, or table area in the conversation or the generated output, and describe the next steps or modification requirements in natural language. The selection defines the scope of action, and MiMo Desktop only processes the part the user wants to modify, eliminating the need to regenerate the complete deliverable or restate the task context.

{/* feishu-style:text-align:left */}
Every modification is saved in the version history. Users can compare different drafts and roll back to any previous version at any time.

> **Example: Partial modification in PPT**: Select the target area → describe the modification requirements → regenerate the corresponding part in real time

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/LnIObUPh7oIHruxtk31cSEFNnpf.mp4" controls={true} playsInline={true} title="实时批改.mp4" class="mdx-video" />

### 04 Smart Scheduling: Matching appropriate execution paths for tasks

{/* feishu-style:text-align:left */}
Users do not need to decide which model, Agent, or execution framework to use before starting a task. MiMo Desktop has a built-in "Smart" scheduling mode that automatically evaluates the task type, complexity, delivery requirements and execution cost, and then automatically matches the most suitable execution path for the current task.

- **Task Assessment**: Identify whether a task falls into the scenario of office work, programming, research, design, or a hybrid scenario, and determine the required reasoning depth, tool scope and delivery standards.

- **Model Routing**: Dynamically select between standard models and flagship models. For daily tasks, standard models are prioritized for their faster speed and lower cost; for complex reasoning, long-chain execution and high-quality delivery, flagship models are adopted for higher output quality. The available model versions are subject to the current release.

- **Harness / Agent / Skill Routing** ：Select office, code or research Harness/Agent based on the task type; large-scale tasks can be split into multiple Agents responsible for research, planning, execution and review respectively, and different Skills will be automatically invoked during the execution process to expand capabilities and meet delivery quality requirements.

- **Multi-session collaboration:** Multi-role session teams can be formed on demand. Multiple sessions can communicate and collaborate with each other, and each session has independent memory, workspace and task status while advancing the project.

{/* feishu-style:text-align:left */}
"Smart" scheduling maximizes the reduction of users' mental cost, and automatically makes the optimal choice among quality, speed and cost.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Igy9bu173oeiGBxVB3acPYMqnVe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Igy9bu173oeiGBxVB3acPYMqnVe.png" width="730" height="387" alt="图片" style="margin: 16px auto;" />

### 05 Browser Control: Serving Both as an Information Entry Point and an Execution Environment

{/* feishu-style:text-align:left */}
MiMo Desktop can independently control the computer browser, read browsing pages, and perform interactive operations, including opening web pages, retrieving information, filling in forms, and extracting creatives. The browser is not only a source of information but also serves as part of the task execution environment.

{/* feishu-style:text-align:left */}
Information obtained from the browser can be directly imported into the current task for research, analysis, organization and content creation, maximizing the time saved for users from manual retrieval and cross-app content transfer. When generating web-based outputs, MiMo Desktop also enables users to inspect key interactions via the browser within the same workflow to automatically verify the deliverables.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YPfhblZmnoQPcOxhivwcWCLXnNh.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YPfhblZmnoQPcOxhivwcWCLXnNh.png" width="1292" height="942" alt="图片" style="margin: 16px auto;" />

### 06 Computer Control: Enter Desktop Environment (Overseas Version Only) 

{/* feishu-style:text-align:left */}
For tasks requiring cross-application collaboration, MiMo Desktop can read screen content, operate the keyboard and mouse to open files, verify data, and transfer information across applications. After the execution of key steps, the system will check the results and make corrections or stop the process when necessary.

{/* feishu-style:text-align:left */}
For stable and repetitive processes, users can record a complete operation once via the Record & Replay function, and then let MiMo Desktop automatically re-execute it through natural language instructions.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/CzlabprNYoZjG0x8MmzcGBtrnEe.mp4" controls={true} playsInline={true} title="demo-5.mp4" class="mdx-video" />

### 07 Keep the Cost of Long Tasks Under Control

{/* feishu-style:text-align:left */}
Long tasks not only need to be completed with high quality, but also require strict control over computing costs. MiMo Desktop reduces unnecessary cost expenditures through optimizations in model routing, in-place editing, and context caching mechanisms.

{/* feishu-style:text-align:left */}
The "Smart" scheduling selects an appropriate model based on task complexity; local editing only processes and regenerates the content that needs modification; the high cache hit optimization minimizes the redundant computation of cached parts to the greatest extent. The combination of these three mechanisms reduces the token consumption caused by model selection, change scope, and redundant context computation respectively.

{/* feishu-style:text-align:left */}
In practical long-task testing, **the cache hit rate can reach up to 99% for the same session and 95% across sessions,** helping users maximize control over task costs.

### Get an Early Access to the New Model

{/* feishu-style:text-align:left */}
During the invitation-based testing period, approved applicants can experience two new-generation Xiaomi MiMo model Preview versions for free on a time-limited and quota-limited basis on MiMo Desktop, codenames:

- MiMo-X-Pro-Preview

- MiMo-X-Flash-Preview

{/* feishu-style:text-align:left */}
The MiMo-X series models are optimized for real professional work scenarios, with key coverage including complex task reasoning, multi-Agent collaboration, large-scale programming projects, Office, web design, front-end and Interaction Design, video editing, 3D generation, music generation, computer control, etc.

{/* feishu-style:text-align:left */}
The MiMo-X series is currently in the preview testing phase, and its performance may change with version iterations. User feedback based on actual tasks will be used for subsequent evaluation and improvement.

### How to join the invitation-only test

{/* feishu-style:text-align:left */}
Visit the following page and submit an application. Once approved, you will be able to download and use MiMo Desktop:

- Overseas: [https://mimo-ai.xiaomimimo.com/desktop/invite/](https://mimo-ai.xiaomimimo.com/desktop/invite/)

- Domestic: [https://mimo.xiaomimimo.com/desktop/invite/](https://mimo.xiaomimimo.com/desktop/invite/)

{/* feishu-style:text-align:left */}
This invitation-only beta will be prioritized for existing users of the Xiaomi MiMo Open Platform.

### 

### Closing Remarks

{/* feishu-style:text-align:left */}
We firmly believe that the significance of AGI lies not only in understanding the world, but more importantly in empowering every individual to create the world.

{/* feishu-style:text-align:left */}
MiMo Desktop is the very first step for us to step into real-world workflows and help users turn their ideas into high-quality deliverables.

{/* feishu-style:text-align:left */}
You are welcome to apply to participate in the invitation-only test, and use real tasks to tell us what it is already capable of accomplishing, and which aspects still need improvement.


--- DOCUMENT: Notice on the Extension of the Limited-Time Experience of MiMo-V2.5-Pro-UltraSpeed ---
URL: https://mimo.mi.com/static/docs/news/latest/beta-extended.md

# Notice on the Extension of the Limited-Time Experience of MiMo-V2.5-Pro-UltraSpeed

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

The limited-time trial (internal testing) of MiMo-V2.5-Pro-UltraSpeed has been successfully concluded. Thank you to every participant. The official commercial version will be launched soon, so stay tuned.

</div>

{/* feishu-style:text-align:left */}
Thank you all for your attention and support for Xiaomi MiMo-V2.5-Pro-UltraSpeed. **After evaluation by the team, we have decided to extend the limited-time experience window for UltraSpeed to allow more developers to have the opportunity to access and experience the inference capability of 1000 tokens/s.** 

{/* feishu-style:text-align:left */}
<br></br>

{/* feishu-style:text-align:left */}
We firmly believe that the ultimate inference speed will bring about brand-new usage scenarios and paradigms in the industry. 

{/* feishu-style:text-align:left */}
Since the launch of MiMo-V2.5-Pro-UltraSpeed, we have received enthusiastic applications from developers across various industries.

{/* feishu-style:text-align:left */}
As of June 23, we have received over 66,000 usage applications from numerous Fortune Global 500 companies, leading enterprises in various industries, and individual developers, covering multiple fields such as law, finance, communications, logistics, automobile manufacturing, culture and media, and higher education institutions, far exceeding our expectations. 

{/* feishu-style:text-align:left */}
<br></br>

{/* feishu-style:text-align:left */}
Seeing the urgent and genuine demand from the community for MiMo-V2.5-Pro-UltraSpeed, we hope to provide more users with the opportunity to experience the paradigm shift brought by 1000 tokens/s, so we have decided **to extend the duration of the chat experience and API access experience for MiMo-V2.5-Pro-UltraSpeed**. 

{/* feishu-style:text-align:left */}
After this extension, you can continue to apply for the closed beta of MiMo-V2.5-Pro-UltraSpeed, and users who have been approved can continue to use it. 

{/* feishu-style:text-align:left */}
The specific offline time will be arranged separately based on resource availability, and we will inform you in advance to ensure that everyone has sufficient time to prepare for migration and adaptation. 

## The application and experience methods remain unchanged 

<table>
<colgroup>
<col style="width: 34.52%" />
<col style="width: 65.47999999999999%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Project</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Description</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Application Entry**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">[https://platform.xiaomimimo.com/ultraspeed](https://platform.xiaomimimo.com/ultraspeed)</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**API Experience**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Continue with the current limited-time trial price (3 times the price of MiMo-V2.5-Pro), with output speed approximately 10 times faster: https://mimo.mi.com/models/en-US/mimo-v2.5-pro-ultraspeed</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Chat Experience**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Users who have passed the review can enjoy a limited-time free chat experience. The entry for the experience is: https://ultraspeed.xiaomimimo.com</span></td>
</tr>
</tbody>
</table>

## Contact Us

{/* feishu-style:text-align:left */}
If you have large-scale commercial use requirements for the UltraSpeed model, please feel free to contact **business-mimo@xiaomi.com** for exclusive support.


--- DOCUMENT: Xiaomi MiMo Claw Official Release | Flagship Model + Kingsoft Office Ecosystem, New Subscription Service Launched  ---
URL: https://mimo.mi.com/static/docs/news/latest/mimoclaw.md

# Xiaomi MiMo Claw Official Release | Flagship Model + Kingsoft Office Ecosystem, New Subscription Service Launched 

{/* feishu-style:text-align:left */}
Xiaomi MiMo Claw is a lightweight cloud-based Claw product, enabling users to quickly experience the MiMo model and access AI Agents with low barriers to entry. Now officially launched, it is powered by **MiMo-V2.5-Pro**, a flagship model deeply optimized for the official OpenClaw framework.

{/* feishu-style:text-align:left */}
We have partnered with Kingsoft Office to deliver a full-suite document productivity solution, added multiple core practical features of OpenClaw, and upgraded benefits for free users: the daily single-session usage limit has been extended. Meanwhile, tiered TokenPlan subscriptions are released for heavy and professional users to cater to diverse usage demands.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/HP03bfQJHokUwcxeptVcNz2on2c.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/HP03bfQJHokUwcxeptVcNz2on2c.png" width="1280" height="723" alt="图片" style="margin: 16px auto;" />

### I. Feature Upgrades

#### 1. Empowered by MiMo Flagship Model for Revolutionary Agent Capabilities

- **Native Protocol Compatibility**: As a pre-built official model for OpenClaw, MiMo-V2.5-Pro natively supports the MCP tool calling protocol and built-in semantic skill parsing, requiring no extra prompt engineering.

- **Massive Long Context Support**: Equipped with an advanced long-context memory scheduling architecture, it handles complex multi-step tasks and enables over 1,000 consecutive tool calls per session, preventing context loss and logical disconnection.

- **Three-Tier MTP Decoding Architecture**: Optimized exclusively for the OpenClaw framework. In standard Agent workflow tests, the overall task reasoning throughput has increased by approximately 3 times.

#### 2. Integrated with Kingsoft Office for One-Stop Document Workflow

- **Wide Format Compatibility**: Supports Word, Excel, PPT, PDF and over 95% of mainstream document formats, covering study, office work, data analysis and other scenarios.

- **End-to-End Workflow**: Seamlessly integrates AI generation, high-definition online preview and real-time editing. No third-party redirection is needed to boost overall office efficiency.

- **Lightweight & User-Friendly**: Users can create content via custom commands or generate standard documents in one click using built-in templates.

#### 3. Improved Computing Efficiency & Lower Token Costs

{/* feishu-style:text-align:left */}
In the ClawEval benchmark test, MiMo-V2.5-Pro achieves a task pass rate (Pass³) of 63.8%. Thanks to superior token utilization, it cuts token consumption by 40% to 60% compared with peer products while delivering equivalent performance. It is fully compatible with all tiers of TokenPlan subscriptions to ensure stable computing power.

#### 4. Cloud Deployment & 24/7 Dedicated Service

{/* feishu-style:text-align:left */}
MiMo Claw runs entirely on the cloud, eliminating local environment configuration and hardware resource occupation. The dedicated personal Agent operates around the clock, supporting background task resumption and automatic error correction to avoid interruptions caused by local device issues.

{/* feishu-style:text-align:left */}
It also incorporates practical functions including multi-channel message interaction, file management, multi-task parallel processing and automatic exception fixing, greatly lowering the learning curve for AI Agent operations.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/LdMdbOilFopR42xppm7cvwZanlb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/LdMdbOilFopR42xppm7cvwZanlb.png" width="740" height="419" alt="图片" style="margin: 16px auto;" />

### II. User Benefits

#### 1. Upgraded Benefits for Free Users

- New users get instant free access, with the daily single-session usage duration extended.

- All core features are unlocked for free accounts, fully meeting the demands of casual daily use and basic office work.

#### 2. Benefits for Existing Users

- **Bundle Upgrade**: MiMo Claw privileges can be added to existing TokenPlan plans. Rights and fees are calculated independently, and the billing cycle will be reset upon order activation.

- **Service Cancellation**: Users who have added MiMo Claw to their plans may cancel the service. Remaining value will be refunded based on unused time.

- **Universal Historical Quotas**: All promotional token quotas from past campaigns are fully usable.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Ev8wbiXPXokCycxWQiKcvE4YnHf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/raw/Ev8wbiXPXokCycxWQiKcvE4YnHf.png" width="785" height="444" alt="图片" style="margin: 16px auto;" />

### III. Subscription Services

{/* feishu-style:text-align:left */}
We have launched tiered TokenPlan subscriptions for users with high-frequency usage and higher computing power requirements.

#### 1. Full Plan Compatibility & Flexible Add-Ons

{/* feishu-style:text-align:left */}
As an independent Agent service, MiMo Claw can be flexibly added to all monthly and annual TokenPlan plans (Lite / Standard / Pro / Max). Existing users can enable the service directly while retaining their original privileges.

#### 2. Dual Subscription Entries for Flexible Purchase

{/* feishu-style:text-align:left */}
Two independent entry points are available for easier subscription:

- **TokenPlan Portal**: When subscribing to TokenPlan on MiMo Platform, users may check the MiMo Claw option to activate the service in one click, with computing quotas and advanced Agent privileges taking effect immediately.

- **Dedicated Claw Portal**: TokenPlan subscriptions on MiMo Studio include MiMo Claw by default, with fixed pricing for each tier.

#### 3. Limited-Time Subscription Pricing

{/* feishu-style:text-align:left */}
Exclusive limited-time offers are available for monthly and annual MiMo Claw subscriptions:

- Introductory Offer: ¥14.9 per month

- Standard Discount: ¥19.9 per month / ¥233.8 per year

{/* feishu-style:text-align:left */}
*Note: Overseas subscriptions are not yet available. Please stay tuned.*

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/PP5Cb6VB0ohcQZxXAf1cg8RlnYg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/PP5Cb6VB0ohcQZxXAf1cg8RlnYg.png" width="724" height="484" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:center */}
**Xiaomi MiMo Platform** 

{/* feishu-style:text-align:center */}
June 2026


--- DOCUMENT: MiMo Code Released and Open-Sourced | Model Agent Collaborative Optimization, Stepping Towards the Self-Evolution Era ---
URL: https://mimo.mi.com/static/docs/news/latest/mimocode.md

# MiMo Code Released and Open-Sourced | Model Agent Collaborative Optimization, Stepping Towards the Self-Evolution Era

{/* feishu-style:text-align:left */}
Today, we officially release and open source **MiMoCode V0.1.0** — an exploratory AI programming assistant running in the terminal.

{/* feishu-style:text-align:left */}
MiMo Code **starts with programming but goes beyond it** . It is not just a useful AI Coding tool, but also an AI teammate who lives in your computer and understands you better the more you use it. 

{/* feishu-style:text-align:left */}
It comes with a built-in top-tier MultiModal Machine Learning model that is free for a limited time  **MiMo-V2.5,**  whose performance is comparable to Claude Sonnet 4.6; at the same time, it supports the integration of mainstream models such as DeepSeek, Kimi, and GLM, as well as third-party Token Plans, to meet the needs of different developers. 

{/* feishu-style:text-align:left */}
MiMo Code is based on the open-source project **OpenCode** for secondary development, **released and open-sourced under the MIT License** . 

##   Core Competencies 

{/* feishu-style:text-align:left */}
**Persistent Memory System + Infinite Context: Solving "AI Amnesia" at the Root**

{/* feishu-style:text-align:left */}
MiMo Code incorporates a built-in and original persistent memory system, which uses three mechanisms - project memory, conversation checkpoints, and task progress - to address the long-standing challenge of "forgetting as you use" in long conversations. Even in long conversations spanning hundreds of rounds, it can maintain output quality and retain key information.

{/* feishu-style:text-align:left */}
Most mainstream Code Agents (Claude Code, Codex, etc.) mostly operate on the principle of "letting AI take its own notes", but the model does not initiate this action on its own; whether and when to take notes depends entirely on its self-awareness. So we changed our approach: **Let the main agent focus on its tasks, and outsource the record-keeping entirely**  - an independent subagent automatically saves the state, and when the window is almost full, it creates a clean summary, allowing the main agent to continue working instead of starting from scratch. 

{/* feishu-style:text-align:left */}
One sentence: **Don't rely solely on the model's self-awareness; use engineering to support it.** 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/RMetbv6RWouvkJxIfP5ck9aKnKe.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/RMetbv6RWouvkJxIfP5ck9aKnKe.jpeg" width="1080" height="553" alt="图片" style="margin: 16px 0;" />

{/* feishu-style:text-align:left */}
**Model Agent Collaborative Optimization + Original Compose Mode**

{/* feishu-style:text-align:left */}
Most Coding Agents work in a way that is like "getting a requirement and then burying themselves in writing code", just like a driver who sets off without looking at the navigation - seemingly efficient, but actually prone to going off track.

{/* feishu-style:text-align:left */}
However, models are not all the same: different models each have their own "personalities" and "endowments", and there are also natural differences in the level of "compatibility" with different Agent frameworks. Simply combining models and frameworks often fails to bring out their true capabilities.

{/* feishu-style:text-align:left */}
MiMo Code has specifically designed a dedicated Harness system for the MiMo series of models, enabling the capabilities of the models to be deeply integrated with the framework; when combined with the original **Compose mode,** it achieves a synergistic effect of 1+1>2. 

{/* feishu-style:text-align:left */}
When using it, simply press the **Tab key** to switch to Compose mode, give it a simple idea, and it can automatically complete the entire process of design, planning, coding, testing, and review, ultimately delivering an industrial-grade finished product.

{/* feishu-style:text-align:left */}
**Actual measurement comparison**

{/* feishu-style:text-align:left */}
We gave the same instructions to both tools:

{/* feishu-style:text-align:left */}
"Help me implement a Redis using Golang, which needs to support connection via redis-cli." 

{/* feishu-style:text-align:left */}
**Claude Code**moves quickly, and the code runs out soon—but there are almost no accompanying tests. The functionality works but is not robust enough, and the risk of subsequent rework is quite high.

{/* feishu-style:text-align:left */}
**MiMoCode** uses the Compose pattern, which initially took more time for planning and seemed to be a bit "slower"; however, in terms of results, it has achieved more comprehensive functionality and is accompanied by complete and detailed testing, truly demonstrating what industrial-grade code should look like.

{/* feishu-style:text-align:left */}
Interestingly,**MiMoCode is actually faster when it comes to calculating the general ledger:** it spends time thinking things through in the early stage and verifying them steadily in the later stage - slow writing, fast verification, resulting in a more worry-free overall experience.

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/Irf9bL6Psoyy1Yxa5TzcXHWEnYg.mp4" controls={true} playsInline={true} title="6112.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
**Dream: Memory settles, the more you use it, the better it understands you**

{/* feishu-style:text-align:left */}
MiMoCode has a built-in unique `/dream `command. It is automatically triggered every 7 days, with an independent Agent reading historical conversations and existing memory files, performing merging, deduplication, validating path effectiveness, and compression, converging scattered memories into a compact current state, and updating the global memory.

{/* feishu-style:text-align:left */}
By the next use, it will automatically recall these memories at the appropriate time. This means that MiMo Code does not start from scratch every time, but continues to grow with an understanding of you and your project—truly becoming more user-friendly the more you use it.

{/* feishu-style:text-align:left */}
**Supports voice input: "A gentleman uses his words, not his hands."** 

{/* feishu-style:text-align:left */}
MiMo Code comes with built-in voice input and control capabilities, powered by the robust speech recognition capabilities of MiMo-V2.5-ASR. Just speak, and the work gets done.

{/* feishu-style:text-align:left */}
What it can do is not just "reading out the prompt": you can verbally correct a wrongly written instruction, or directly issue operation commands such as "send" or "execute" - from input to control, without touching the keyboard throughout the process, and efficiency naturally takes another step up. 

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/Adfpb7ISwokvbax1IgLcjpT4nOf.mp4" controls={true} playsInline={true} title="6113.mp4" class="mdx-video" />

##  Let Data Do the Talking: Same Model, Stronger Performance

{/* feishu-style:text-align:left */}
On two authoritative test sets **SWE-Bench** and **Terminal Bench** that target real-world programming scenarios, we conducted a set of Controlled Experiments: we let MiMo Code and Claude Code **use the same MiMo model,** and only compare their respective Agent systems themselves. 

{/* feishu-style:text-align:left */}
Results show that MiMo Code achieved **62% (** Claude Code achieved 57%) on SWE-Bench Pro, and **73% (** Claude Code achieved 68%) on Terminal Bench 2 —— on the premise that the models are exactly the same, MiMo Code obtained a better score through the synergy of its exclusive Harness and Compose mode.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Ki8Bb4voCosxwTxqbhScDg6ynTe.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Ki8Bb4voCosxwTxqbhScDg6ynTe.jpeg" width="1080" height="365" alt="图片" style="margin: 16px 0;" />

##   How to Use: Zero Configuration, Out Of The Box 

{/* feishu-style:text-align:left */}
**Installation and Startup:** Open the terminal

- Recommended for Mac and Linux users: `curl -fsSL https://mimo.xiaomi.com/install | bash`

- Windows users are recommended to use npm:`npm install -g @mimo-ai/cli`

{/* feishu-style:text-align:left */}
After installation, enter ` mimo  in the terminal ` to start. For the best experience, it is highly recommended  **Mac users to use it in iTerm or the vscode terminal.** 

- **Model Configuration**

   - Built-in MiMo-V2.5 Limited-Time**Free Channel,** Available Without Registration

   - Compatible with mainstream model APIs such as DeepSeek / Kimi / GLM, as well as third-party Token Plans 

- **Usage:** 

   - Input `/ ` View Each Configuration 

   - All settings are fully Chinese localized, making it locally friendly

   - The right side of the TUI page has a permanent status dashboard for observing work progress at any time 

{/* feishu-style:text-align:left */}
For more technical details, please follow our team Blog:https://mimo.xiaomi.com/mimocode

##   Open Source and Outlook 

{/* feishu-style:text-align:left */}
MiMo Code **was released and open-sourced** , under the permissive MIT License — which means it is open to almost everyone: 

- **Individual developers** can freely use, modify, and distribute, and do whatever they want;

- **Enterprises**  can integrate it into their own development toolchains without worrying about licensing constraints; 

- **The community**  can build vertical programming assistants based on it, giving rise to more possibilities. 

{/* feishu-style:text-align:left */}
We believe that the value of a good tool lies not only in its functionality but also in how many people can contribute to its refinement and where it is headed. We look forward to working with you to make MiMo Code even better.


--- DOCUMENT: Xiaomi MiMo Partners with TileRT | 1T Model Breaks 1000 tokens/s Output Speed for the First Time ---
URL: https://mimo.mi.com/static/docs/news/latest/1000tps.md

# Xiaomi MiMo Partners with TileRT | 1T Model Breaks 1000 tokens/s Output Speed for the First Time

<div className='mdx-highlight mdx-highlight-info' data-highlight-icon='info'>

The limited-time trial (internal testing) of MiMo-V2.5-Pro-UltraSpeed has been successfully concluded. Thank you to every participant. The official commercial version will be launched soon, so stay tuned.

</div>

{/* feishu-style:text-align:left */}
**Xiaomi MiMo-V2.5-Pro-UltraSpeed: In all martial arts under heaven, only speed prevails**

{/* feishu-style:text-align:left */}
From the first sports car of the internal combustion engine era to supersonic flight breaking the sound barrier, humanity's thirst for speed is ingrained in our genes. 

{/* feishu-style:text-align:left */}
The speed of AI inference also determines the boundaries of intelligence. When the model is fast enough, it is no longer just a tool that "waits for results" but becomes an extension of your thinking - real-time response, instant iteration, and seamless collaboration. 

{/* feishu-style:text-align:left */}
**Today, MiMo × TileRT jointly announced: The UltraSpeed mode of Xiaomi MiMo-V2.5-Pro enables the output speed of flagship models with trillions of parameters to exceed 1000 tokens/s for the first time.** 

{/* feishu-style:text-align:left */}
How fast is 1000 tokens/s exactly:

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/NZOzbnChDoMHyPxxKc0cOvfUnwf.mp4" controls={true} playsInline={true} title="6123.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
 **Limited-time opening · Application-based experience**

{/* feishu-style:text-align:left */}
MiMo-V2.5-Pro-UltraSpeed API is now available for synchronization, offering a limited-time trial price, priced at 3 times that of MiMo-V2.5-Pro, while providing approximately 10 times the output speed improvement!**3 times the price increase, 10 times the output experience.** (Only API experience is supported, Token Plan is not supported for the time being)

{/* feishu-style:text-align:left */}
This MiMo-V2.5-Pro-UltraSpeed will be open for a limited time on an application basis. Users whose applications are approved can access the API for a limited time to experience it, with the time limited from June 9 to June 23, 2026, 23:59. 

{/* feishu-style:text-align:left */}
**Within 12 hours after the release of MiMo-V2.5-Pro-UltraSpeed on the X (Twitter) platform last night, more than 3,000 enterprises and their developers applied for trial access to the UltraSpeed API, with industry categories covering law, finance, hospitality, telecommunications, logistics, internet, culture and media, automobile manufacturing, etc.** 

{/* feishu-style:text-align:left */}
👉 **Application Entry:** https://platform.xiaomimimo.com/ultraspeed

{/* feishu-style:text-align:left */}
Due to limited inference resources, the number of trial slots for this time is limited. After submitting an application, we do not guarantee the timeliness of review or the review pass rate; we will prioritize the review of enterprises and professional developer scenarios with real business needs. If you have large-scale commercial needs related to UltraSpeed, please feel free to contact business-mimo@xiaomi.com.

{/* feishu-style:text-align:left */}
**💬 Chat Experience (Limited-time Free)** 

{/* feishu-style:text-align:left */}
Users who pass the review can also obtain**time-limited free**Chat experience (valid within a two-week open window), with the experience entry:https://ultraspeed.xiaomimimo.com

{/* feishu-style:text-align:left */}
To ensure the quality of experience and fairness of use under resource-constrained conditions, the trial service rules are as follows: Each account can successfully enter the queue up to 10 times per day; the maximum duration of a single session is 30 minutes; if a session remains idle for more than 5 minutes, the system will automatically release the resources. 

##  1000 tokens/s, not only fast, but also a qualitative change in paradigm

{/* feishu-style:text-align:left */}
On the scale of trillion parameters (1T), breaking through 1000 tps is not just about making the typewriter faster; it brings about a fundamental subversion of the AI application paradigm.

{/* feishu-style:text-align:left */}
**First, speed itself begins to transform into intelligence.**  In the past, when faced with difficult problems, you could only "wait for an answer and pray it was correct"; now, within the same waiting time, the model can run dozens of inference paths in parallel (Best-of-N/Tree Search), automatically verify and correct errors in the background, and use "speed" to derive the depth of thinking, directly improving the quality of inference.

{/* feishu-style:text-align:left */}
**Secondly, it has completely liberated the productivity limit of Coding Agent.** Previously, when asking AI to write code, developers were often confined by the inference speed and had to sit in front of the screen waiting painfully. However, the ultra-fast inference of 1000 tps has brought about a revolutionary code writing speed and a soaring production efficiency.

{/* feishu-style:text-align:left */}
**Most importantly, trillion-scale models are starting to enter the real-time decision-making closed loop.**  The millisecond-level "thinking-response" cycle enables the 1T flagship model to seamlessly integrate into time-sensitive scenarios - high-frequency quantitative trading signal generation, instantaneous anti-fraud risk control interception, intelligent bidding, and real-time interactive dialogue. 

{/* feishu-style:text-align:left */}
When this power is injected into life - critical scenarios such as surgical assistance and medical image analysis,  **the speed of AI is no longer just a measure of efficiency, but a bargaining chip in a race against death.**  In the operating room, for every second that AI completes lesion analysis and risk prediction ahead of time, the room for doctors to act increases. This makes us deeply convinced that the ultimate significance of speed is not just to boost productivity, but to enable technology to provide a better life for humanity. 

##  Optimal Codesign of Model and System

{/* feishu-style:text-align:left */}
Achieving an output speed of over 1000 tokens/s for the 1T flagship model is not a breakthrough in a single technology, but rather the result of in-depth collaboration between the MiMo model and TileRT system teams and their pursuit of ultimate codesign. When the current industry pursues similar extreme speeds, it often chooses to take the dedicated hardware route, such as Cerebras' wafer-scale integration (Wafer-Scale) or Groq's custom chip architecture based on pure on-chip SRAM. We chose to implement even more impressive inference speed on  **general-purpose GPUs**  through model-system co-design. 

{/* feishu-style:text-align:left */}
On the model side, in response to the bandwidth bottleneck of general-purpose hardware, **FP4 quantization** was performed, significantly reducing the model size and memory access overhead; meanwhile, an efficient speculative decoding **DFlash** based on block-level masked parallel prediction was introduced, greatly increasing the Token acceptance length per validation. On the system side, TileRT is perfectly adapted to the dynamic characteristics of the algorithm, customizing a specially optimized compilation engine and compute kernel for the new quantization and speculative decoding processes.**Thanks to this ultimate Codesign, we were able to enable a 1T model to exceed an output speed of 1000 tokens/s using only a standard 8-card general-purpose GPU node.** 

### FP4 Quantization

{/* feishu-style:text-align:left */}
At the scale of trillion parameters (1T), traditional 8-bit (such as FP8/INT8) or even 16-bit inference can lead to extremely high GPU memory usage and memory bandwidth pressure. Reducing the bit width of parameters can directly contribute to the decoding output speed. Therefore, we adopted the industry-standard and almost lossless FP4 (MXFP4[1]) quantization. 

{/* feishu-style:text-align:left */}
However, if FP4 quantization is applied uniformly to the entire model, the accuracy and generalization ability of the model in complex reasoning and logical code often degrade. Given the typical MoE (Mixture of Experts) architecture characteristics of Xiaomi MiMo-V2.5-Pro, its Experts account for the vast majority of the parameters and have the highest tolerance for quantization accuracy. Therefore, we choose to perform parameter FP4 quantization only on the MoE Experts while retaining the original accuracy for other modules. Through such FP4 QAT (Quantization-Aware Training), we have significantly reduced the model size and maximized hardware bandwidth utilization while keeping the overall capabilities of the model basically on par with the original model, as shown in the figure below: 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BQLkb9B9no6fCsxAGtXcdGLynsd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BQLkb9B9no6fCsxAGtXcdGLynsd.png" width="790" height="290" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:center */}
 [FP4 vs FP8 Model Comparison]

### DFlash Speculative Decoding

{/* feishu-style:text-align:left */}
Traditional Speculative Decoding relies on a small draft model to "guess" subsequent tokens, which are then verified by a large model. This approach transforms the autoregressive generation that produces 1 token per forward pass into parallel generation of multiple tokens, and the rejection sampling mechanism in the large model verification process ensures that the output quality remains intact.However, its bottleneck lies in the fact that the quality of the draft model determines the acceptance rate, while a stronger draft model brings higher computational overhead, making it difficult to achieve both.

{/* feishu-style:text-align:left */}
To break this deadlock, we adopted the innovative **DFlash** block-level masked parallel prediction method [2] in academia: the draft model fills an entire block of masked positions simultaneously in a single forward pass, fundamentally removing the serial constraint of "draft autoregression".

{/* feishu-style:text-align:left */}
We have implemented and customized this path on MiMo-V2.5-Pro, targeting trillion-scale MoE and long context scenarios. Through the Muon second-order optimizer and model self-distillation, we ensure that small mask blocks can still provide an ideal acceptance rate while compressing the overhead of the draft stage to near the limit: 

- The Draft model fully adopts the Sliding Window Attention (SWA) mechanism, which is naturally aligned with the SWA design of the MiMo-V2 series models. This enables Draft to no longer rely on a complete prefix, and the computing power for a single prediction changes from linearly increasing with the context length to a constant level. 

- During training, the mask signal sampling is offloaded to the local GPU sharding, enabling a single sequence step to generate tens of thousands of independent training signals covering different lengths of context positions, aligning with the long context capabilities of the MiMo-V2 series models while avoiding cross-device communication overhead. 

{/* feishu-style:text-align:left */}
In terms of effectiveness, our parallel prediction speculative decoding has achieved a significant increase in acceptance length in multiple agent and high-value coding scenarios, meaning that the large model can "confirm" more content in one go during each verification; in addition, we limit the mask block size to 8 to reduce verification overhead and improve concurrency, enabling the high acceptance length to be directly translated into high inference throughput: 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Xe7QbGXAzoyN0uxkEAEckok8nsh.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Xe7QbGXAzoyN0uxkEAEckok8nsh.png" width="758" height="338" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:center */}
[Acceptable Length of DFlash in Different Scenarios]

{/* feishu-style:text-align:left */}
In the Coding scenario, the average acceptance length can reach 6.30, and in some samples, it reaches a maximum of 7.14. This means that among the 8 draft tokens per round of verification, 6-7 tokens can be accepted, and the draft pushes the acceptance rate to a level where end-to-end truly benefits while maintaining lightness. We also find that in the general conversation scenario with more divergent semantics and higher uncertainty, the current acceptance rate is not yet high, and we are continuously optimizing the algorithm to explore a higher generalization ceiling. 

### TileRT Ultra-Low Latency Inference Kernel/System

{/* feishu-style:text-align:left */}
If the algorithmic reconstruction of MiMo has removed the heavy bandwidth shackles for the trillion and quadrillion models, then the TileRT inference system has directly squeezed the physical potential of general-purpose GPUs to the absolute limit of the microsecond level. 

{/* feishu-style:text-align:left */}
Under the ultra-high frequency operating state of 1000 tokens/s, the lifecycle of a single operator is compressed to the microsecond level, and the "operator boundary" of traditional inference systems has become the core bottleneck - each operator launch, hardware synchronization, and global memory round-trip will interrupt the entire execution flow on the microsecond scale, exposing an obvious "Execution Gap". 

{/* feishu-style:text-align:left */}
**Paradigm-level Execution Model Transformation of TileRT**As the underlying infrastructure for ultra-low latency inference, TileRT introduces a brand-new execution model that fundamentally eliminates the execution gaps caused by operator boundaries:

- **Persistent Engine Kernel**: Completely abandons the traditional operator-by-operator startup mode, allowing the entire computing pipeline to reside permanently inside the GPU for continuous operation. This enables the system to achieve whole-link continuous prefetching capabilities. While the current tile is still being computed in the Tensor Core, subsequent data has already flowed in advance along the storage architecture, achieving the ultimate overlap of data transfer and computation.

- **Heterogeneous Pipeline Collaboration (Warp Specialization)** : At the Tile level, communication, data movement, and tensor computation are more finely physically disassembled. By breaking the original homogeneous serial pace, different Warps (thread bundles) and even the heterogeneous execution domains of the entire GPU can each perform their own duties and collaborate precisely, completely evolving the GPU into a continuously flowing and precisely coordinated heterogeneous execution system. 

{/* feishu-style:text-align:left */}
**Deep convergence of software and hardware (Codesign) at the microsecond scale**When the underlying execution model pushes hardware performance to its limits, pure runtime optimization begins to hit physical limitations. On this basis, the TileRT system team and Xiaomi MiMo team have carried out in-depth technical co-creation, breaking the original software hierarchical barriers.To ensure that the model behavior perfectly aligns with the continuous advancement of this ultra-low latency execution pipeline, the model layer ultimately adopted an FP4 hybrid quantization strategy targeting MoE Experts and implemented DFlash speculative decoding aligned with SWA on the trillion-scale architecture. TileRT closely coordinated with these algorithmic features and quantization schemes, customizing the underlying compilation engine and compute kernels. Both parties made profound joint engineering trade-offs based on hardware physical limitations, enabling the execution pressure to ultimately achieve a smooth closed-loop within the hardware boundaries.

{/* feishu-style:text-align:left */}
The emergence of 1000 tokens/s is by no means a coincidence of single-point optimization. It is an inevitable result of the deep convergence and co-evolution of high-level system infrastructure and cutting-edge algorithm models towards each other. 

{/* feishu-style:text-align:left */}
The TileRT team is a cutting-edge system architecture team focused on next-generation AI infrastructure and dedicated to achieving ultra-low latency inference. 

{/* feishu-style:text-align:left */}
The team is committed to driving millisecond-level real-time response of cutting-edge large models in production environments, breaking traditional storage and computing barriers with a brand-new runtime architecture. The team has deduced and implemented a brand-new paradigm-level execution model. 

{/* feishu-style:text-align:left */}
Through full stack breakthroughs in underlying technologies such as persistent kernels, tile pipelines, and heterogeneous collaboration, TileRT has achieved maximum computing power release in the complex heterogeneous computing ecosystem. 

{/* feishu-style:text-align:left */}
As an enabler of core infrastructure, the team actively collaborates with top industry partners to conduct co-design of software and hardware, building a high-performance computing power foundation for the autonomous era that is extremely eager for "ultimate speed". 

{/* feishu-style:text-align:left */}
For more technical details on TileRT, please read:https://www.tilert.ai/blog/breaking-1000-tps-zh.html

##   More Effect Displays 

{/* feishu-style:text-align:left */}
Create a Snake game in just 10 seconds

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/Q3YMbSc2rovooTx8g2mcLjAGn0Y.mp4" controls={true} playsInline={true} title="6124.mp4" class="mdx-video" />

{/* feishu-style:text-align:left */}
Replicate a MacOS system in just 1 minute

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/DM63bOuVLo7T9Uxs1tic17rpnHg.mp4" controls={true} playsInline={true} title="6126.mp4" class="mdx-video" />

##  Open Source and Outlook

- We have open-sourced the MiMo-V2.5-Pro-FP4-DFlash checkpoint to HuggingFace, including FP4 quantization weights and DFlash model parameters 

- Welcome the community to use and provide feedback:https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash

- The ultimate inference support for MiMo-V2.5 is on the way, stay tuned! 

{/* feishu-style:text-align:left */}
*MiMo × TileRT, the ultimate co-design of model and system, enables trillion-parameter models to achieve an ultimate inference speed of 1000 TPS.*

{/* feishu-style:text-align:left */}
[1]https://www.opencompute.org/documents/ocp-microscaling-formats-mx-v1-0-spec-final-pdf

{/* feishu-style:text-align:left */}
[2]https://arxiv.org/abs/2602.06036


--- DOCUMENT: MiMo-V2.5 Series Inference Full-Link Optimization: Pushing Hybrid SWA Efficiency to the Extreme ---
URL: https://mimo.mi.com/static/docs/news/latest/updqate.md

# MiMo-V2.5 Series Inference Full-Link Optimization: Pushing Hybrid SWA Efficiency to the Extreme

{/* feishu-style:text-align:left */}
The MiMo V2.5 series models (including MiMo-V2.5, MiMo-V2.5-Pro, etc.) integrate multiple architectural features: Hybrid Sliding Window Attention (Hybrid SWA) reduces KVCache storage to approximately 1/7 of Full Attention through hybrid window attention; MoE reduces the computational cost per token while maintaining model capacity through sparse activation; MultiModal Machine Learning Encoder supports cross-modal understanding of vision, audio, video, etc. The combination of the three endows the MiMo-V2.5 series of models with significant potential for effectiveness and efficiency in long-context and multi-modal scenarios.

{/* feishu-style:text-align:left */}
From the very beginning of the design of the MiMo-V2 model, we have had a very clear goal: to train a model that is both sufficiently powerful and efficient in long text reasoning scenarios. However, these two goals inherently have tension in engineering. Strong reasoning ability means that the model needs to have the ability to model long-range dependencies, which usually corresponds to larger-scale attention computation and higher KVCache overhead. In the traditional Full Attention architecture, both the computational complexity of Attention and the storage of KVCache grow rapidly with the context length, making the cost of long context training and inference quickly become unacceptable. The core idea of Hybrid SWA is to perform hierarchical mixing between local window attention (Sliding Window Attention, SWA) and global attention (Full Attention): the vast majority of layers only compute attention within local windows, and only a small number of key layers retain a global view. Theoretically, this structure can reduce the computational complexity of Attention to nearly linear while still maintaining the ability to model long-range dependencies. 

{/* feishu-style:text-align:left */}
However, the theoretical architectural advantages do not naturally translate into efficiency advantages in real online systems. On the one hand, Hybrid SWA significantly increases the complexity of KVCache cache hits, prefix matching, and the maintenance of Full / SWA semantic consistency. On the other hand, in real engineering systems, data transfer in multi-level storage, inconsistencies between asynchronous prefetch and scheduling, and the difficulty of aligning distributed cache states together make it difficult to directly realize the theoretical benefits.

{/* feishu-style:text-align:left */}
In addition to Hybrid SWA, MoE places higher demands on distributed scheduling and load balance; the throughput bottleneck of MultiModal Machine Learning Encoder in large graph and long video scenarios also urgently needs to be broken through. Moreover, the optimization of general components such as scheduling strategies and Prefill/Decode execution links is also indispensable. What this article records is precisely a whole-link engineering practice of an inference system centered around the MiMo-V2.5 series of models, covering KVCache management, hierarchical cache systems, SWA prefix cache trees, scheduling strategies, Prefill/Decode execution links, and MultiModal Machine Learning optimization, systematically translating the efficiency potential of the architecture (especially Hybrid SWA) into the production environment.

## 1. Inference Efficiency Advantages of the Hybrid SWA Architecture

{/* feishu-style:text-align:left */}
Before delving into specific discussions on optimization work, it is necessary to first quantify the efficiency upper bound of Hybrid SWA - this serves not only as the starting point for architecture selection but also as the theoretical benchmark and theoretical upper bound for all subsequent system optimizations. 

### 1.1 Computational Complexity Analysis

{/* feishu-style:text-align:left */}
Taking MiMo-V2.5-Pro as an example, this model has a total of 70 layers, among which 10 layers are Full Attention, the remaining 60 layers are SWA, and the sliding window size of SWA is 128. Compared with Full Attention, the computational complexity of Hybrid SWA is shown in the figure. Since the proportion of SWA layers is 6/7, the computational complexity of the Hybrid SWA architecture is approximately 1/7 that of Full Attention. In the Chunk Prefill scenario, Prefill is approximately compute-bound, and this gap is basically equivalent to the theoretical reduction in Prefill cost. 

### 1.2 KVCache Storage Analysis

{/* feishu-style:text-align:left */}
Since the SWA layer only needs to retain KV within the sliding window and does not need to store the full sequence, the KVCache occupancy also drops to nearly 1/7. The Decode phase is approximately memory-bandwidth-bound, and its latency is proportional to the read volume of model parameters plus KVCache. In the case of long sequences, the volume of KVCache may far exceed the model parameters, so the reduction in KVCache storage is almost directly equivalent to the reduction of decode cost in long sequence scenarios. 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Q8rLbWiIwoiUpYxK7opcwghbnac.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Q8rLbWiIwoiUpYxK7opcwghbnac.png" width="584" height="389" alt="图片" style="margin: 16px auto;" />

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ZGpBby5AlopLvnxgycwc5ZYlnsr.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ZGpBby5AlopLvnxgycwc5ZYlnsr.png" width="582" height="388" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
There are significant differences in the KVCache storage of different model architectures, and there are also differences in memory access patterns. The following is an estimate of the KVCache size for each domestic model. It can be seen that MiMo-V2.5-Pro and MiMo-V2.5 rank second among domestic models in terms of KVCache, second only to DeepSeek-V4-Pro and Flash. 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YZfzb97DrohQ9Yx0dTZcK7T9njf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YZfzb97DrohQ9Yx0dTZcK7T9njf.png" width="588" height="392" alt="图片" style="margin: 16px auto;" />

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/HC6hb5QdMocBz1xYJGJcwp7vnXc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/HC6hb5QdMocBz1xYJGJcwp7vnXc.png" width="570" height="380" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
It should be noted that the actual cost difference is not strictly equivalent to the KVCache scale ratio, because there are also fixed computational and memory access overheads that are independent of the sequence length. However, in the long context scenario, the overall trend remains consistent:  **the cost-effectiveness of short texts is similar, and the longer the sequence, the greater the inference cost advantage** . 

## 2. KVCache System Refactoring

{/* feishu-style:text-align:left */}
As models that were among the first to adopt the Hybrid SWA architecture, the MiMo-V2 and MiMo-V2.5 series faced the issue that, at the time, neither mainstream open-source inference frameworks nor caching systems provided complete support for SWA. At the beginning of the MiMo API launch, we chose SGLang v0.5.5 as the codebase for the service backend. Immediately afterwards, we encountered a severe test. In the then version, SGLang's HiCache did not support SWA, or rather, the early SWA support was implemented in a way that "compatible with SWA at the cost of storing Full KVCache". Although there are some temporary solutions to make SWA more usable, we still hope to develop a KVCache system with a higher ceiling and better usability. 

### 2.1 SWA KVCache Management

#### KVCache Dual Pool 

{/* feishu-style:text-align:left */}
Hybrid SWA introduces a core storage contradiction: the Full Attention layer needs to retain full-sequence KV (O(N)), while the SWA layer only needs to maintain KV within the sliding window (O(W)). However, under the traditional single KV pool design, the system must uniformly allocate video memory for all layers according to O(N), preventing the window sparsity of SWA from being utilized, and the actual storage efficiency degrades to an approximate implementation of Full KVCache.

{/* feishu-style:text-align:left */}
To address this issue, a natural approach is to split KVCache into two independent pools, Full Attention and SWA, and perform unified abstraction at the system level: 

- **Physical Level**: Maintain Full KV pool and SWA KV pool separately. The SWA pool only configures its capacity according to the window size and supports independent eviction based on the window, thereby strictly limiting SWA storage to the O(W) scale; this mechanism also extends to the L2 and L3 storage levels.

- **Logical Level**: Still exposes a single sequence view to the upper layer (prefix tree, scheduler, and transport protocol), with the Full Attention index as the authoritative index, and maintains the mapping relationship from Full to SWA to achieve transparent hierarchical storage.

- **Scheduling Constraints** : When the system receives an access request, it simultaneously verifies the capacity constraints of Full KV and SWA KV to avoid misjudgment of resources in a single dimension. 

- **Data Transfer**: Cross-layer transfer is performed solely based on the SWA mask to ensure that only data within the valid window is transferred, thereby avoiding redundant bandwidth occupation.

{/* feishu-style:text-align:left */}
Through the above design, SWA KVCache achieves strict O(W) storage constraints at the system level, increasing the overall KVCache capacity efficiency by approximately 7×, thereby truly unleashing the structural advantages of Hybrid SWA. Mainstream inference frameworks have also adopted similar implementation schemes. 

#### KVCache asynchronous layer-by-layer pulling 

{/* feishu-style:text-align:left */}
After the implementation of SWA KVCache storage optimization, the SWA layer only needs to prefetch a very small amount of KVCache, which enables the process of prefetching KVCache from the Host to the Device to achieve perfect overlap through layerwise granularity scheduling. As a result, the cost of Cache reads during the inference process is close to zero. 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/VRLjb2butogonJx8uKwcnlIrnpf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/VRLjb2butogonJx8uKwcnlIrnpf.png" width="570" height="328" alt="图片" style="margin: 16px auto;" />

#### SWA-aware prefix cache tree 

{/* feishu-style:text-align:left */}
The hit rule of traditional RadixAttention is based on a simple assumption:  **equal token sequences → equal KV** . This assumption holds true in Full Attention mode - as long as two requests share the same token id, their corresponding KV must still be in the pool and can be directly reused. 

{/* feishu-style:text-align:left */}
However, this assumption is broken in SWA mode. The reason is that the logical lifecycle of the prefix tree does not align with the physical lifecycle of SWA KV. The length of prefix tree nodes is not constrained by the SWA window. The sequence length of a node can be either shorter or much longer than the window; moreover, nodes change continuously with request merging, splitting, and removal. Thus, although a prefix tree node still logically represents a complete token sequence, its corresponding SWA KV  **may only have its last part remaining, or may even have completely disappeared.** If the prefix tree still gives the reuse length according to the rule of "matching when tokens are equal", what the scheduler receives may be a pseudo-match where "the tail KV has evaporated" - subsequent attention calculations will read invalid or overwritten slots, directly affecting the model's performance. 

{/* feishu-style:text-align:left */}
To ensure that prefix reuse remains correct and efficient in SWA mode, it is necessary to modify the semantics of the prefix tree in three aspects: 

1. **The matching rule has been upgraded to "window safety length"** : In addition to token equality, it is also necessary to ensure that at least W tokens at the end still have valid slots in the SWA pool. The matching length is trimmed to this new boundary - the part beyond it is treated as a miss. This ensures that all KV retrieved from the hit segment must be valid. 

1. **The elimination path is bound to the request lifecycle** : Each chunk of long prefill completion, request end, and decode generating a certain number of tokens all trigger a release of out-of-window SWA. This ensures that the SWA pool occupancy remains constant at the W or chunk level in long context/long output tasks, rather than growing with the sequence length. 

1. **The node simultaneously carries two sets of indexes** : Each prefix tree node records two pieces of information - the Full Attention segment index (determines the logical order and participates in the Full Attention layer computation) and the SWA segment mapping (determines window security). During eviction, they also need to be managed separately: the SWA segments outside the window can be evicted individually while retaining the Full Attention segments (so that the prefix can still be reused by the Full Attention layer), or the entire segment can be evicted. 

{/* feishu-style:text-align:left */}
SWA compressing the KV volume to 1/7 is  **the gain at the capacity level** , while the hit rate is  **the gain at the reuse level** . Only when the two are multiplied can we obtain the curve of the actual computational cost during the prefill phase. After introducing the "window safety length" matching rule, the hit rate of KVCache with the same token capacity theoretically decreases slightly, but the number of tokens under the same storage capacity reaches several times, resulting in a substantial increase in the actual hit rate. 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Aqkfbfnpro65cxxaq1lcG16Qn4d.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Aqkfbfnpro65cxxaq1lcG16Qn4d.png" width="544" height="418" alt="图片" style="margin: 16px auto;" />

#### Optimization of KVCache Hit Rate Improvement

{/* feishu-style:text-align:left */}
After all three levels of HiCache are transformed into SWA-aware, the device, host, and storage backend each maintain a set of states indicating "which locations have valid SWA". However, HiCache's data transfer link is asynchronous, cross-deployment caches vary, and the length of shared prefixes across sessions also varies - which means that inconsistencies are likely to occur between the Full Attention Cache and valid SWA indices on each end. According to the matching rules of the SWA-aware prefix cache tree, if a sequence can be hit on the Full Attention Cache but not on the SWA Cache, it will lead to severe matching length truncation. The more truncation there is, the longer the length that needs to be recalculated, and the lower the optimization effect of the SWA Cache. Therefore, we have optimized distributed consistency and cache hit rate for different scenarios:

- **Device side complete, host side missing**: When L3→L2 prefetching only pulls in the tail segment of the sequence due to bandwidth and latency trade-offs, or when the L1 prefix tree is not synchronized to L2/L3 after reorganization, the problem of a complete device side but a missing host side may occur.In response, we proactively check the difference set of the occupancy of the two sets of SWA (device/host) at time points such as node merging in the prefix tree and completion of prefill, allocate additional slots in the SWA pool on the host side, and asynchronously write the SWA KV of the device from D2H. 

- **Host side is complete, device side is missing**: Wait until the next H2D completes natural alignment, no need for active patching.

- **High-frequency sequence L3 prefix expiration**: The head of the long sequence persists in L1/L2 due to high-frequency access, and Cache affinity routes requests with the same prefix to the same node. As a result, the Cache in L3 may be cleaned up by the storage eviction policy due to long-term lack of direct access, leading to the premature release of the L3 Cache of the global high-frequency sequence and a significant reduction in cross-machine reuse. To address this, when accessing the Cache on L1/L2, we query the L3 Cache at a certain interval to avoid eviction.

- **SWA Retention Strategy for Medium and Short Sequences** : For SWA of medium and short sequences, we fix relatively dense SWA KV Caches at certain lengths based on user request patterns. Although increasing the density of SWA will raise the proportion of SWA in the overall KV Cache, it can directly benefit scenarios such as multi-user sharing of system prompts. 

{/* feishu-style:text-align:left */}
Through the above optimizations, we have truly transformed the expansion at the KV Cache capacity level into a high effective hit length, making it possible to reuse long prefixes across sessions, which is particularly beneficial for scenarios such as long conversations of agents, multi-user sharing of system prompts, and tool calls with multiple accesses to the same codebase. 

### 2.2 GCache: High-performance Distributed Cache Infrastructure

{/* feishu-style:text-align:left */}
GCache is a high-performance general-purpose cache developed by Xiaomi's Storage Team, and it is an important part of building the "training and inference integrated" storage system. In the early days of the training scenario, the Storage Team realized that some open-source cache projects had limited acceleration effects on distributed file systems and could not fully unleash their performance, so they embarked on the path of self-developed. Later, with the release of the MiMo large model and the launch of inference services, the team also transformed GCache into an independent storage product for model distribution and as the L3 KVCache of the inference engine. 

{/* feishu-style:text-align:left */}
GCache supports both file and KV semantics, multi-level caching of memory/disk/remote, has shm memory persistence and whole-link zero-copy capabilities, supports advanced features such as high-concurrency non-blocking IO and RDMA communication, meets the performance requirements of upper-layer services for high throughput and low latency, and has good scalability. 

#### Architecture Design 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OXC0bNUtKo9kSRxeLBdcQ6Gsnfh.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OXC0bNUtKo9kSRxeLBdcQ6Gsnfh.png" width="562" height="309" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
GCache has several characteristics: 

1. **The decentralized metadata management approach allows the cluster scale to expand without limitations:** 

   - Calculate the consistent hash for the key to determine the storage location. 

   - Master uses Raft high-availability deployment. However, Master is only responsible for managing heartbeats and Service Discovery, and the IO path does not pass through Master.

1. **The server supports both memory and disk caching simultaneously:** 

   - Cold data in memory will be evicted to disk, while hot data on disk will be promoted to memory. This mode is very friendly to inference scenarios, automatically ensuring the performance of active sessions and reducing the cost of sessions that have not been started for a long time. 

   - Memory supports persistence to shm, ensuring that cached data is not lost when the service is restarted.

   - Supports smooth scaling up or down of machines without losing cache during the process. 

1. **Provide multi-language SDK, start a dedicated thread, slice and dispatch user requests:** 

   - Does not occupy the resources of user threads; slicing improves concurrency and controls the IO size within a range friendly to RDMA.

   - The thread callback operates in Asynchronous Mode, allowing flexible control of callback granularity, such as single kv level, batch level, or CUDA stream level. 

#### Network Optimization

{/* feishu-style:text-align:left */}
Currently, mainstream GPU models are all equipped with eight 400G high-performance network cards. However, even when considering the separate deployment of PD, the current inference frameworks still struggle to fully utilize the network bandwidth, leading to voices in the industry calling for reducing the configuration of network cards to cut costs.

{/* feishu-style:text-align:left */}
To fully leverage the capabilities of high-speed networks, GCache preferentially uses GPU network cards instead of front-end network cards for communication, and has made extensive optimizations on the communication module, such as NUMA binding and same-track affinity. In terms of specific performance metrics, when using 1MB-sized IOs, the RDMA read throughput of a single process can reach 170 GB/s, with a latency of only 280 us; in the GDR scenario, due to the higher bandwidth of HBM, a single process can reach approximately 350 GB/s, which is sufficient to meet the communication performance requirements of inference frameworks.

#### Storage Cost Optimization

{/* feishu-style:text-align:left */}
2026 will undoubtedly be a year when the industry is highly sensitive to storage costs. Unlike other competitors that use dedicated storage models, GCache preferentially adopts the method of co-locating on GPU machines, taking over part of the memory of Prefill and Decode nodes, as well as several NVMe SSDs that come with the machine, with additional storage costs being 0.

#### Stability Assurance

{/* feishu-style:text-align:left */}
Due to the mixed fabric, the high failure rate of GPU machines has become a hurdle in front of stability. Since its launch, GCache has basically encountered machine failures where the server is located every day. First, the team spent a great deal of time and effort strengthening the fault handling logic of the code; second, since the keys are fully scattered by consistent hashing, by performing pre-logical grouping on session IDs, the fault radius has also been reduced; finally, by combining the hardware detection capabilities provided by the underlying platform, faults are detected in advance, and automated processes are used for data migration. For extremely rare sudden downtime events that cannot be addressed in advance, by setting a lower SDK timeout, the inference framework can promptly detect misses and perform recalculation, ensuring that the front-end inference process is not significantly affected. 

{/* feishu-style:text-align:left */}
Based on the above work, GCache has been able to maintain single-copy storage in the hybrid deployment state without using multi-copy means to improve availability, which is also one of the important reasons for the relatively low storage cost.

### 2.3 Discussion on Cache Hit Rate

{/* feishu-style:text-align:left */}
Thanks to the aforementioned optimizations for SWA on KVCache - lower storage footprint, supplemented by a more stable large-capacity GCache as L3 storage - we have been able to significantly extend the TTL (Time-To-Live) of the cache, thereby substantially increasing the hit rate of the KV Cache. The eviction of KVCache essentially stems from storage capacity constraints. When the capacity approaches saturation, the system must prioritize retaining the KV Cache generated by new requests and evict historically accessed entries according to strategies such as LRU, which directly results in the fact that a certain context often fails to be hit when reused after several hours. The extremely small storage footprint of SWA enables the cache capacity for concurrent requests that can be supported to increase exponentially under the same cost, while the large-capacity L3 further expands the available capacity at low cost—the more abundant the storage space, the less pressure there is for KVCache to be evicted, and naturally the longer its retention duration.The longer the TTL, the wider the hit window for historical context, and the cache hit rate rises accordingly. Additionally, although the smaller bandwidth transmission pressure of SWA does not directly affect TTL, it significantly reduces the data transfer overhead between multi-level storage, providing a guarantee for the stable and efficient operation of the entire cache system. 

{/* feishu-style:text-align:left */}
Since the model went live, we have continuously observed on the server side that under the mainstream high-quality harness framework, the server-side KV Cache hit rate can reach an average of  **93%** ; for individual users with high-intensity and long-term usage, this indicator can even climb to  **95%**  or higher. In the future, we will continue to iterate on the KV Cache management logic of SWA and collaborate with more harness frameworks to promote the co-design of harness-inference to further optimize the upper limit of the cache hit rate. 

## III. Scheduling Optimization

{/* feishu-style:text-align:left */}
In the early days of the SGLang community, the router service was not yet fully mature, and there was no data sharing among multiple instances. If a router service unexpectedly fails or requests are routed to different router services, the issue of KVCache scheduling fallback will occur. To address this issue and ensure high availability in large-scale Clustered Deployment, Xiaomi developed the dynamically scalable stateless scheduler LLM-Router. By using Redis as a centralized storage, it avoids the KVCache scheduling fallback phenomenon after a single-service failure, thereby ensuring a more stable cache hit rate. 

### 3.1 KVCache and Load Affinity Scheduling

{/* feishu-style:text-align:left */}
Since HiCache is highly sensitive to the L2 hit rate, if there is a miss in L2, it is necessary to search in L3 and fetch the KVCache, and only after the fetching is completed can the request be inferred. Increasing the L2 hit rate from the router side can reduce unnecessary synchronous waiting, thereby directly improving throughput performance.

{/* feishu-style:text-align:left */}
Router implements KVCache affinity scheduling by maintaining distributed requests in a Radix prefix tree. It preferentially selects among multiple Prefill instances**nodes that have cached the prefix of the current request**, and at the same time**takes load balance into account**to avoid hot spot skew. After the strategy was launched, it increased the cache hit rate of L2 by approximately**25%** , and the single-machine input throughput by approximately**30%** . Its core formula is roughly:

```bash

# 选择 score 最大的 worker，含义：缓存命中率高 + 负载低 = 得分高 = 优先选择score(worker) = matchWeight × prefix_match_percentage − normalized_load
```

### 3.2 TTFT Optimization

{/* feishu-style:text-align:left */}
When queuing occurs in model services, the traditional First Come First Serve strategy does not consider the priority relationship between requests with high and low hit rates, causing requests with more cache hits but fewer real computational tokens to potentially wait for requests with lower cache hit rates to finish inference before they can start. The TTFT P99 of the overall service becomes extremely long, slowing down the average performance of the overall throughput. 

{/* feishu-style:text-align:left */}
To address this issue, when the waiting queue selects to prioritize the execution of the prefill service, the Router side preferentially schedules requests with fewer real computational tokens, avoiding the problem of P99 degradation caused by blocking requests that originally had short computation times. Meanwhile, this strategy may lead to the starvation phenomenon where some requests are not scheduled for a long time, so we have also added a waiting time penalty mechanism to balance this phenomenon. The results show that this strategy does not degrade the service quality for shorter requests, while for longer requests, it can reduce the P90 metric of TTFT by up to  **30%** . 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/QzALbfqBbobRUIxKJMucTWDAned.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/QzALbfqBbobRUIxKJMucTWDAned.png" width="568" height="405" alt="图片" style="margin: 16px auto;" />

## 4. Prefill Optimization

### 4.1 Distributed Configuration

{/* feishu-style:text-align:left */}
Theoretically, the smaller the EP (Expert Parallelism) during the prefill phase, the better the performance and throughput, which is mainly reflected in three aspects: fewer machines are involved, resulting in lower cross-machine communication overhead; fewer DP (Data Parallelism) numbers, reducing the impact of attention load differences between DPs on performance; and each machine can carry more experts, leading to better MoE load balance. However, the size of EP is constrained by video memory and needs to meet the video memory requirements of model parameters and KVCache. Early SWA KVCache needed to store the KVCache of all tokens, resulting in an excessively large EP; after optimization, only the KVCache of partial SWA tokens needs to be stored, and we have reduced the EP to 1/2 of its original size.**End-to-end performance has improved by approximately 40%.** Subsequently, we will continue to explore PP (Pipeline Parallelism) optimization for the Hybrid SWA architecture to further reduce the EP size and improve overall throughput.

### 4.2 Length Bucketing Strategy

{/* feishu-style:text-align:left */}
Compared to the pure GQA architecture, the Hybrid architecture of the MiMo-V2.5 series significantly improves computational efficiency, but throughput still decreases significantly as the sequence length increases. The following figure shows the throughput in Chunked Prefill when computing 16K tokens with different prefix lengths:

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/B6DubYIBroLfFwx3dwAcHH4Rndh.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/B6DubYIBroLfFwx3dwAcHH4Rndh.png" width="540" height="360" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
In the Agentic scenario, most ultra-long requests originate from multi-round agent interactions and generally carry a large amount of prefix cache. When requests with significantly different lengths are scheduled to the same model instance, short requests will be dragged down by long requests, reducing overall throughput, which is mainly reflected in two scenarios: 

1. **DP-Attention Synchronization**: Multiple DPs need to perform collective communication after each layer of attention computation to synchronize the computation entering the MoE stage. If both long and short requests exist simultaneously on multiple DPs within the same EP group, the short requests will be slowed down by the computation of the long requests;

1. **Chunked Prefill Interference**: When requests with different prefix lengths are grouped into the same chunk for inference, short-prefix requests are slowed down by the computation of long-prefix requests.

{/* feishu-style:text-align:left */}
To alleviate the above load imbalance issue, we adopted**the three-level length bucketing strategy**(0–64K / 64K–256K / 256K–1M), aggregating requests with similar load characteristics into the same bucket for computation,**significantly improving the average throughput of online prefill**. On this basis, we are currently exploring a more fine-grained and flexible bucketing mechanism to adapt to the dynamic changes of online loads.

### 4.3 MoE Load Balance

{/* feishu-style:text-align:left */}
All models in the MiMo-V2.5 series adopt the MoE architecture, and the issue of expert load balance during the prefill phase needs to be considered. Since a load balance training objective was introduced during the pre-training phase and the training was relatively stable, the model has learned a relatively uniform expert allocation strategy during training. During the inference phase, without enabling any expert load balance strategy, **the average expert load per layer (the ratio of the average number of tokens across all ranks in a layer to the maximum number of tokens in a rank of that layer) is approximately 0.85, which is already at a relatively optimal distribution level**. Therefore, we currently have not introduced any expert load balance strategy. Subsequently, we will continuously monitor this indicator and introduce relevant optimizations as needed based on the dynamic evolution of the online load pattern.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ZlGnbF1JWoujQ6xYx51c0UCMn1c.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ZlGnbF1JWoujQ6xYx51c0UCMn1c.png" width="594" height="297" alt="图片" style="margin: 16px auto;" />

### 4.4 Resolve NUMA Conflicts

{/* feishu-style:text-align:left */}
In some Ubuntu systems, the kernel numa_balancing parameter conflicts with the numa-node configuration of SGLang, resulting in occasional large execution gaps between computing kernels during model inference. In a multi-node multi-GPU deployment, the occurrence positions of these gaps on each rank are random, and each time synchronization occurs between ranks, the overall computing process is slowed down by the slowest rank, thereby significantly affecting the overall inference efficiency. After disabling the numa_balancing parameter of the system kernel, the issue was resolved,  **and end-to-end performance increased by approximately 10%** . 

## 5. Decode Optimization

### 5.1 Video Memory Optimization

{/* feishu-style:text-align:left */}
In the Agentic scenario, multi-round conversations cause the context to continuously grow, and the video memory occupancy of KVCache has become the main bottleneck in decoding. After the video memory is fully occupied by KVCache, the batch size cannot be expanded, resulting in underutilized computation and limited decoding throughput. As a result, we can only increase the number of nodes to handle concurrency, which drives up inference costs. To increase the concurrency of a single node, we have implemented multiple video memory optimization measures: 

1. **Decode KVCache fully supports SWA**: The effective capacity of KVCache has increased by nearly 5 times**.** 

1. **Optimization of KVCache Preallocation in PD Separation**: Move the prealloc process of requests that have not yet started from GPU memory to CPU memory, and only move it into GPU memory when the decode actually starts, eliminating waste caused by resource preoccupation.

1. **CUDA Graph Memory Optimization**: Optimize CUDA Graph parameters to reduce space waste and increase available memory.

### 5.2 MTP Optimization

{/* feishu-style:text-align:left */}
The MiMo-V2.5 series models natively support 3-layer MTP accelerated decode output, but MTP was not enabled during the prefill phase previously, resulting in the MTP using invalid KVCache for the initial 128 output tokens of decode and a very low prediction acceptance rate. Since most output sequences are short in Agentic scenarios, this defect significantly reduces the effectiveness of MTP acceleration. By introducing MTP support during the prefill phase and performing specialized adaptation and optimization on HiCache L2/L3, the acceleration effect of MTP in the early stage of decoding has been significantly improved:  **The acceleration ratio reaches 2.3× for the 0–128th tokens and 1.5× for the 128–256th tokens** , effectively reducing the real decoding cost in agentic scenarios. 

## 6. MultiModal Machine Learning Inference Optimization

{/* feishu-style:text-align:left */}
Based on the SGLang Community v0.5.7 EPD solution, we have carried out**a large number of engineering optimizations and stability fixes in EPD separation**around MiMo-V2.5, increasing the Encoder throughput to **2 times**while keeping the latency unchanged. Currently, we are feeding back the work results to the SGLang Community (issues#24945). The differences in Encoder performance before and after optimization are shown in the following table:

<table>
<colgroup>
<col style="width: 25%" />
<col style="width: 25%" />
<col style="width: 25%" />
<col style="width: 25%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">QPS</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Avg Latency</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">P90 Latency</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Before Optimization</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">15</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">78.39ms</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">100.76ms</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">Optimized</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">30</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">80.28ms</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">82.94ms</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
**6.1 Architecture Optimization**

- **MultiModal Machine Learning Embedding data replication and inference parallel** : In the Prefill scheduler main loop, support MultiModal Machine Learning Embedding data **asynchronous replication** between TPs, overlap with Prefill inference, and reduce GPU idle time.

- **Encoder supports Data Parallelism**: Since the Encoder model is relatively small, setting TP>1 will actually lead to performance degradation. Therefore, we deploy Encoder with TP=1 and at the same time **support Data Parallelism**, which facilitates deployment and operation and maintenance on a single machine with 8 GPUs.

- **Encoder supports cross-request group batching**: We introduced cross-request batching for the EPD Encoder Server, aggregating concurrent requests by modality through the Encoder scheduler, where the image/audio of multiple requests **is merged into a single Encoder Forward** and then split and returned according to the requests, thus solving the problem of low GPU utilization in per-request encoding.

### 6.2 Preprocessing Optimization

- **GPU Preprocessing of Images**: In large image scenarios, performing resize/normalize/patchify on the CPU will significantly increase end-to-end latency. Therefore, we move the preprocessing to the GPU,**eliminating the CPU bottleneck**. 

- **Parallel Image Download**: We use **multi-process parallelism** to download images and perform PIL decoding, avoiding the latency caused by sequential downloading and GIL lock contention.

- **MultiModal Machine Learning Download and Forward Parallelism**: In the early Encoder implementation, both data download and inference between batches and within batches were sequential, resulting in the GPU being idle during the download; we decoupled download and inference through a message queue to achieve**overlap between download and inference within batches**.

- **Video parallel decoding**: evenly divide the frame extraction indices into N chunks, and create an independent VideoDecoder instance for each chunk to perform multi-threaded parallel decoding, reducing the end-to-end latency of the 1-hour video Encoder **from 156s to 23s**.

### 6.3 Cache Optimization

- **Encoder Consistent Hashing**: In scenarios with multiple Encoders, Prefill uses polling to select Encoders, resulting in a decrease in the MultiModal Machine Learning cache hit rate. Through consistent hashing, we route the same requests to the same Encoder, **increasing the cache hit rate by 30%** . 

- **In-machine Embedding Cache Sharing** : Through shared memory, multi-modal cached data is shared among multiple cards within the Encoder machine, improving cache hit rate. 

## VII. Postscript

{/* feishu-style:text-align:left */}
Looking back at the entire project, the inference efficiency of the MiMo-V2.5 series models does not stem from a single-point breakthrough in one particular stage, but rather from the result of multi-dimensional collaborative optimization. Hybrid SWA benefits both prefill and decode, but the under-optimized implementation of KVCache actually increases costs at each stage.Around this goal, we systematically reconstructed KVCache management, hierarchical caching, and prefix cache trees, tackled the core issues of SWA KVCache, optimized the scheduling strategy and Prefill/Decode links, and after verification in real online scenarios, finally translated its theoretical efficiency advantages into actual production environments. Only then did Hybrid SWA demonstrate its architectural advantages of combining strength and efficiency in  **long text inference** . By further combining MoE configuration and various optimizations of MultiModal Machine Learning inference, the performance of online inference services has been significantly improved. 

{/* feishu-style:text-align:left */}
We hereby present the first large-scale engineering implementation plan that comprehensively covers the Hybrid SWA + MoE + MultiModal Machine Learning combined architecture, and will pass on the cost savings achieved thereby to users through API price cuts. Meanwhile, we have already contributed some optimizations to the SGLang open-source community in the form of PRs, and will continue to advance more open-source initiatives, hoping to ensure that engineering optimization no longer serves as a barrier as soon as possible, enabling this type of composite architecture that combines strength and efficiency to be more widely explored and applied.


--- DOCUMENT: Xiaomi MiMo Orbit’s first batch of Agent ecosystem co-building partners announced ---
URL: https://mimo.mi.com/static/docs/news/latest/v2.5-orbit.md

# Xiaomi MiMo Orbit’s first batch of Agent ecosystem co-building partners announced

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/CUG2bZX5XoEtjVxSCSIcTOJmnhf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/raw/CUG2bZX5XoEtjVxSCSIcTOJmnhf.png" width="510" height="1645" alt="图片" style="margin: 16px auto;" />


--- DOCUMENT: MiMo-V2.5 Series Price Adjustment Announcement | 100 Trillion Token Creator Incentive Plan Concludes ---
URL: https://mimo.mi.com/static/docs/news/latest/v2.5-price-update.md

# MiMo-V2.5 Series Price Adjustment Announcement | 100 Trillion Token Creator Incentive Plan Concludes

{/* feishu-style:text-align:left */}
Over the past few months, through activities such as MiMo Orbit and the Quadrillion Token Creator Incentive Program, we have enabled more people to experience MiMo and solve real problems - this is the first step for MiMo on the path to large-scale application.

{/* feishu-style:text-align:left */}
Now, with the continuous improvement of underlying technologies, we can finally do something more thorough - **permanently renovate the entire model pricing system**. 

{/* feishu-style:text-align:left */}
**Quick Overview of the Core of This Announcement:** 

- MiMo-V2.5 Series API Permanent Price Reduction

- Token Plan billing system optimization, with usage increased to 5-8 times the original

- The Creator Incentive Program for Quadrillion Tokens Concludes Successfully 

- Full reset of the current effective Token Plan user quota

{/* feishu-style:text-align:left */}
Effective Time: 0:00, May 27, 2026, Beijing Time

## MiMo-V2.5 Series API Permanent Price Reduction

{/* feishu-style:text-align:left */}
Compared to the original API pricing, the new pricing can have a maximum reduction of up to 99%, and no longer differentiates based on the input length.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/KxrVbPwOMofb91xhtzucgLxBnVg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/KxrVbPwOMofb91xhtzucgLxBnVg.png" width="516" height="880" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
**This price adjustment officially takes effect at 0:00 on May 27th, Beijing time, with global synchronization. We sincerely invite all developers to integrate and experience it.** 

## Optimization of TokenPlan Billing System

- **Increase the quantity without increasing the price, with the usage volume increased to 5-8 times the original, unlocking more abundant productivity for you** 

   - **Example: In Agent or Code scenarios, the number of available tokens is:** 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AflpbOEmtomRQEx3oGZc2myHnrg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AflpbOEmtomRQEx3oGZc2myHnrg.png" width="503" height="604" alt="图片" style="margin: 16px auto;" />

- **Billing rules have been adjusted to be clearer, more understandable, and what you see is what you get.** 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Q04gb754RoR9G8xmgfQcOWlFnCc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Q04gb754RoR9G8xmgfQcOWlFnCc.png" width="525" height="878" alt="图片" style="margin: 16px auto;" />

## The Creator Incentive Program for Quadrillion Tokens Concluded Successfully 

{/* feishu-style:text-align:left */}
Since its launch on April 28, the "Trillion Token Creator Incentive Program" has been enthusiastically pursued and widely followed by users worldwide. As of 16:08 on May 26, Beijing Time, all 100T Tokens have been fully distributed ahead of schedule, and the event has concluded successfully ahead of schedule. We thank all developers for their enthusiastic participation! 

{/* feishu-style:text-align:left */}
Note: The exclusive welfare activities for members of the Apache Software Foundation are valid for a long term, can continue to be applied for, and are not affected by this finalization.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BJYObcw4foA2MMxFj4ZcsW66nZb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BJYObcw4foA2MMxFj4ZcsW66nZb.png" width="435" height="685" alt="图片" style="margin: 16px auto;" />

## Surprise: All existing TokenPlan user quotas have been fully reset

{/* feishu-style:text-align:left */}
Regardless of the current usage of the package, the Credits quota of all users who have subscribed to the Token Plan and are still within the validity period (including users who participated in the Quadrillion Token Creator Incentive Program and obtained the Token Plan, covering users with exclusive benefits from the Apache Software Foundation) will be fully reset at 0:00 on May 27th, Beijing Time, and implemented according to the new billing rules.

{/* feishu-style:text-align:left */}
One More Thing: For historical paid users whose Token Plan has expired, we have also prepared surprise gifts, which will be announced within the next week. Please stay tuned.

## Optimization Instructions for Inference Technology

{/* feishu-style:text-align:left */}
Behind this price adjustment is the continuous optimization of the inference system by Xiaomi's technical team.

{/* feishu-style:text-align:left */}
We fully support SWA (Sliding Window Attention) based on SGLang HiCache, reducing the data transfer volume of KV Cache among multi-level storage such as GPU memory, CPU memory, and SSD to nearly  **1/7**  of that before optimization, and increasing the number of cacheable tokens to nearly  **5 times**  of that before optimization, significantly improving cache hit rate and inference efficiency. 

{/* feishu-style:text-align:left */}
Meanwhile, we further enhanced the input throughput capacity of the cluster by optimizing the expert parallelism scheme, input length bucketing strategy, etc., thereby continuously reducing the service cost per token while ensuring service quality.

## Conclusion

{/* feishu-style:text-align:left */}
The value of technology ultimately lies in the breadth of its use. 

{/* feishu-style:text-align:left */}
Relying on continuous technological innovation, we hope to leverage real, sustainable, and large-scale inference demand by providing model services that combine low cost with top-notch capabilities, thereby promoting the construction of a complete AI infrastructure chain. 

{/* feishu-style:text-align:left */}
Enabling more people to use better models - this is MiMo's unwavering mission.


--- DOCUMENT: Xiaomi MiMo-V2.5 series open-sourced & Orbit 100 trillion token plan launched ---
URL: https://mimo.mi.com/static/docs/news/latest/v2.5-open-sourced.md

# Xiaomi MiMo-V2.5 series open-sourced & Orbit 100 trillion token plan launched

{/* feishu-style:text-align:left */}
Today, we officially open source the Xiaomi MiMo-V2.5 series, which uses the MIT license, supports commercial inference deployment and secondary training, and requires no additional authorization.

## Open protocol, fully open source

{/* feishu-style:text-align:left */}
The MiMo V2.5 series models began public testing on April 23rd. We thank all users for their enthusiastic feedback and encouragement during this period.

{/* feishu-style:text-align:left */}
This series includes two models, both supporting a 1-million-token context window:

- mimo-v2.5-pro: Designed for complex task scenarios, deeply optimized for Agent and Coding applications. It ranks first among open-source models globally on the GDPVal-AA and ClawEval leaderboards.

- mimo-v2.5: A native full-modal model supporting text, image, video, and audio understanding, with powerful Agent capabilities.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/VZxrbdHSUoqx63x5RtycLBnznfe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/VZxrbdHSUoqx63x5RtycLBnznfe.png" width="693" height="324" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
We deeply understand that the true value of a model does not lie in its ranking on leaderboards, but rather in its ability to efficiently assist developers in solving real-world problems. On the Claw-Eval leaderboard, MiMo V2.5 series ranks at the optimal frontier of task completion rate and Token efficiency 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BhnNbzCq5oBDXCxWPOcc9umtnPe.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BhnNbzCq5oBDXCxWPOcc9umtnPe.jpeg" width="582" height="404" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
After undergoing refinement and verification during the public beta phase, this series has further improved in terms of intelligence level and stability, and has reached the standard for release. 

{/* feishu-style:text-align:left */}
Today, we are releasing the model weights of the MiMo V2.5 series to global developers under the MIT License, and at the same time, we are collaborating with chip manufacturers and inference frameworks to provide adaptation code, hoping to contribute to the open-source community and developer ecosystem.

{/* feishu-style:text-align:left */}
The weights of both models (including the Base model) have been fully open-sourced under the permissive MIT license, allowing free commercial use, secondary training, and fine-tuning without additional authorization.

> Model weight collection: [https://huggingface.co/collections/XiaomiMiMo/mimo-v25](https://huggingface.co/collections/XiaomiMiMo/mimo-v25)

{/* feishu-style:text-align:left */}
For more details, refer to the model Blog:

{/* feishu-style:text-align:left */}
https://mimo.xiaomi.com/index#blog

## MiMo Orbit Program

{/* feishu-style:text-align:left */}
We believe that the value of open source lies not only in the public disclosure of weights, but more importantly, in the co-construction of the ecosystem. 

{/* feishu-style:text-align:left */}
To this end, we are officially launching the MiMo Orbit Program. 

{/* feishu-style:text-align:left */}
The MiMo Orbit plan is divided into two parts, namely the " **Creator Trillion Token Incentive Plan"** for AI builders and the " **Agent Ecosystem Co-construction Plan** " for Agent framework teams.

### Creator Trillion Token Incentive Program 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/QzbXbqNIlou0rYxl8NjcZgQKn9f.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/QzbXbqNIlou0rYxl8NjcZgQKn9f.png" width="575" height="269" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
Xiaomi MiMo will distribute free Tokens to global users, with a total of **100 trillion (100T) Tokens** to be distributed within 30 days, and the distribution will end once all Tokens are given out.

{/* feishu-style:text-align:left */}
This event adopts an application system, and users whose applications are approved will receive the Max-tier Token Plan at most, which includes 1.6 billion Credits and is worth 659 yuan. 

{/* feishu-style:text-align:left */}
**Event Time**

{/* feishu-style:text-align:left */}
From 00:00 on April 28, 2026, to 00:00 on May 28, 2026, Beijing Time

{/* feishu-style:text-align:left */}
**Participation Method**

{/* feishu-style:text-align:left */}
You can fill out the application via the following link or QR code. We will carefully evaluate each application material and match corresponding benefits based on your usage scenarios and needs. Successful applicants will receive our follow-up emails. 

{/* feishu-style:text-align:left */}
Application URL: [100t.xiaomimimo.com](http://100t.xiaomimimo.com)

{/* feishu-style:text-align:left */}
Application QR Code: 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ZDwPbXuHSoURfdxMjW7cxqnmnMe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ZDwPbXuHSoURfdxMjW7cxqnmnMe.png" width="238" height="238" alt="图片" style="margin: 16px auto;" />

### Agent Ecosystem Co-construction Initiative

{/* feishu-style:text-align:left */}
Xiaomi MiMo provides specialized support to the global Agent Framework Team. We will offer limited-time free support for the Agent Framework, enabling your users to access and experience the MiMo series of models with zero barriers. 

{/* feishu-style:text-align:left */}
During the model ecosystem adaptation process, we have carried out in-depth cooperation with Agent framework vendors such as OpenCode, Hermes Agent, and KiloCode, and received a great deal of positive feedback and recognition.

<table>
<colgroup>
<col style="width: 25.93%" />
<col style="width: 24.37%" />
<col style="width: 25.21%" />
<col style="width: 24.49000000000001%" />
</colgroup>
<thead>
<tr>
<th><img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BFeBbuvrVoxBXfxmpkPcSSASnCx.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BFeBbuvrVoxBXfxmpkPcSSASnCx.png" width="197" height="192" alt="图片" style="margin: 16px 0;" /></th>
<th><img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/GByLbKMNGoOqutxZcybcY9D8nOt.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/GByLbKMNGoOqutxZcybcY9D8nOt.png" width="198" height="193" alt="图片" style="margin: 16px 0;" /></th>
<th><img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/GjbfbWQbho1sE2xayLIc3TCunSd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/GjbfbWQbho1sE2xayLIc3TCunSd.png" width="192" height="187" alt="图片" style="margin: 16px 0;" /></th>
<th><img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/WlzwbOfDXorn7OxU45UczO0xnae.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/WlzwbOfDXorn7OxU45UczO0xnae.png" width="186" height="181" alt="图片" style="margin: 16px 0;" /></th>
</tr>
</thead>
<tbody>
</tbody>
</table>

{/* feishu-style:text-align:left */}
We welcome like-minded Agent framework developers and manufacturers to contact us：[ business-mimo@xiaomi.com ](mailto:business-mimo@xiaomi.com)

## Chip ecosystem and inference framework adaptation

{/* feishu-style:text-align:left */}
mimo-v2.5-pro completed the integration and adaptation with multiple chip manufacturers on the first day of its open source release. The following is a partial list of manufacturers:

- Ali T-HEAD

> The T-HEAD Zhenwu 810E relies on a full-stack self-developed AI software stack to achieve deep adaptation. 

- Amazon Web Services

> Amazon Web Services (AWS) has completed the in-depth adaptation of mimo-v2.5-pro based on its self-developed Trainium2 chip, Neuron SDK, and vLLM inference framework, achieving first-day adaptation where the model is globally available upon open-sourcing. The next-generation 3nm process Trainium3 will further unleash the Agentic performance potential of the model. 

- AMD

> AMD, relying on the ROCm open-source software stack, provides Day-0 adaptation and comprehensive optimization support for mimo-v2.5-pro, helping developers and enterprise users efficiently complete model deployment and go live. 

- Baidu Kunlun Chip

> Kunlun Chip relies on its self-developed architecture, effectively ensuring the stable and efficient operation of models on the platform through underlying operator optimization and software-hardware co-acceleration, and building a solid computing power foundation for upper-layer applications. 

- Suiyuan Technology

> Suiyuan Technology relies on its self-developed Yusuan TopsRider software stack for in-depth optimization. mimo-v2.5-pro has completed full adaptation on Suiyuan L600, achieving stable operation with high throughput and low latency, and maintaining excellent performance in complex tasks and long sequence scenarios. 

- Muxi 

> Muxi Xiyun C Series relies on the full stack self-developed MXMACA software stack to achieve end-to-end native support from Triton syntax to Muxi GPU instruction set, with better performance. 

- Days Intelligence Chip

> Tianshu Zhixin can achieve Day 0-level deep adaptation of models, relying on full-stack self-developed software and hardware to build high-quality computing power, with efficient adaptation and easy migration, capable of precisely unleashing model performance and ensuring stable operation. 

{/* feishu-style:text-align:left */}
In addition, the MiMo-V2.5 series models have also completed Day-0 adaptation for the mainstream inference frameworks SGLang and vLLM.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/F6upbwMkIol7iex8R0Sc81XYnhh.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/F6upbwMkIol7iex8R0Sc81XYnhh.png" width="397" height="535" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
From the first-generation model to today's full open source of MiMo-V2.5 series, every step of MiMo's growth has been inseparable from the community's feedback and co-construction. 

{/* feishu-style:text-align:left */}
We will continue to invest in the iteration of model capabilities and the improvement of the ecosystem, and work together with global developers to enable Agent to truly enter every application scenario.


--- DOCUMENT: Xiaomi MiMo-V2.5-TTS-Series + ASR Officially Launched: Your Voice, Under Your Control ---
URL: https://mimo.mi.com/static/docs/news/latest/v2.5-tts-release.md

# Xiaomi MiMo-V2.5-TTS-Series + ASR Officially Launched: Your Voice, Under Your Control

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/K1WEbfl2EolOVvxoIWPcgjlMnJh.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/K1WEbfl2EolOVvxoIWPcgjlMnJh.jpeg" width="452" height="452" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
Speech technology is undergoing such a transformation: from "being able to listen and read" to "precise understanding and flexible expression". In real creative and interactive scenarios, machines not only need to penetrate complex spoken language environments - dialect accents, environmental noise, multiple people speaking simultaneously - but also use voice to shape characters, grasp emotions, so that expression is no longer just about conveying words, but also conveying feelings.

{/* feishu-style:text-align:left */}
Whether it's content creators or businesses relying on speech technology, what they truly need is a speech system that can be freely controlled by language: input a noisy meeting recording, and it can accurately transcribe; input a director's note saying "this part should be low and angry", and it can generate a fitting performance. It understands everything and can express everything. 

{/* feishu-style:text-align:left */}
To this end, we officially release today **MiMo-V2.5-TTS Series** and **mimo-v2.5-asr**  — a whole-link speech model series for the Agent era, covering the two core capabilities of recognition and synthesis, enabling both speech input and output to be freely scheduled by language.

- The MiMo-V2.5-TTS Series includes three models, which have now been launched on  **Xiaomi MiMo Open Platform** , and  **are available for free for a limited time** . The three models share unified style instruction following, audio label control, and text understanding capabilities, enabling voice performance to be precisely regulated by language, respectively covering three typical creative needs: 

   - **mimo-v2.5-tts:**  Built-in with multiple high-quality premium voices, supports fine-grained control over speech rate, emotion, tone, etc., Out Of The Box, meeting multi-scenario expression needs.

   - **mimo-v2.5-tts-voicedesign:** Quickly define and generate a brand-new voice in one sentence, making voice creation more intuitive and efficient.

   - **mimo-v2.5-tts-voiceclone:** High-fidelity replication of target timbre with a small number of samples, while maintaining stable style instruction following and audio label control capabilities.

{/* feishu-style:text-align:left */}
**MiMo-Studio Quick Experience Address:**  **https://aistudio.xiaomimimo.com/#/c**

- **mimo-v2.5-asr is officially open-sourced.** The model's speech recognition performance in complex real-world scenarios such as Chinese-English bilingual, Chinese dialects, Code-Switch, strong noise, and multi-speaker has reached the industry-leading level, providing clear and reliable speech transcription for Agents and ensuring that every interaction is based on accurate understanding.

## mimo-v2.5-tts: Let Voice Become Everyone's Creativity

### Core Features of TTS Series

#### Precise ability to follow style instructions

{/* feishu-style:text-align:left */}
From short single-sentence instructions to an entire director's notes, the model can consistently understand and follow them, covering multiple dimensions such as emotion, tone, speaking speed, vocalization style, and language style. Instructions do not need to be written as structured parameters - simply describe the desired feeling as if giving a briefing to an actor, and the model will translate it into the corresponding performance. 

{/* feishu-style:text-align:left */}
For scenarios with higher consistency requirements - such as audio dramas, game NPCs, and character-based dialogues - the model also supports **director script-level** structured input:  **characters** ,  **scenes** ,  **detailed instructions** are described in layers, with each layer independently updated at its own pace and freely combined. This layering not only ensures that the timbre identity of the character remains consistent throughout, but also allows the performance of each sentence to be individually controlled. 

{/* feishu-style:text-align:left */}
**Case1**

{/* feishu-style:text-align:left */}
Instruct :

{/* feishu-style:text-align:left */}
声音低沉沙哑一点，像个历经沧桑的老前辈在讲述传奇人物。语气里带点由衷的敬佩，娓娓道来。

{/* feishu-style:text-align:left */}
Text：

{/* feishu-style:text-align:left */}
街口那个老周啊，媳妇走得早，一个人拉扯俩娃，白天蹬三轮，晚上还去夜市摆摊修鞋。现在俩孩子都有出息喽，想接他去城里享福——他不去，就守着那间小铺子。哎，人哪，骨头硬，心里头就踏实。

{/* feishu-style:text-align:left */}
Audio（Voice name：冰糖）：

<mimo-audio src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/audios/EnS1b1joRoQAzZxTENKcu55QnMg.mp3" controls={true} title="db1c31a7-03ef-41ec-a340-10b8499be584.wav" class="mdx-audio"/>

{/* feishu-style:text-align:left */}
**Case2**

{/* feishu-style:text-align:left */}
Instruct :

```bash
CHARACTER
曾是守护九天的神祇，见证了凡人的无药可救后，决定以灭世来完成最终的净化。他的心中装满悲悯，但手段是绝对的屠戮。

SCENE
悬浮于崩塌的祭坛之上，俯视下方在火海中哀嚎、曾奉他为信仰的信徒。他在降下最后的毁灭前，发出神圣却残忍的叹息。

DIRECTION
发声机制与共鸣：充分打开胸腔共鸣，制造一种神圣的回音感。声音位置靠后，音色如古钟般低沉且带有金属质感的磁性。
声调与韵律：四声（去声）的下落要极其平缓，不要砸实，带有一种吟诵古籍般的从容与宏大。字句之间的停顿拉长，展现出视万物为刍狗的威压。
气声与实声的较量：在说前两句时，实声饱满，高高在上；但在说出“闭上眼吧”时，声音突然混入大量疲惫的气息，神性开始出现裂痕，流露出勉强的残忍。
咬字细节：古风词汇（如“垂怜”、“沉疴”、“剔骨刮毒”）咬字要深，声母起音圆润而不尖锐。结尾的最后半句，几乎全部转化为气声，像是在哄睡一个婴儿，将残酷包裹在极致的悲哀之中。
```

{/* feishu-style:text-align:left */}
Text：

{/* feishu-style:text-align:left */}
你们求我垂怜，求我降下甘霖洗净这浊世。可这世间的沉疴，唯有烈火能剔骨刮毒。闭上眼吧。这业火烧起来的时候，一点也不疼。

{/* feishu-style:text-align:left */}
Audio（Voice name：白桦）：

<mimo-audio src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/audios/VmQkbuCDyo4i59x0gtXcDp1Encd.mp3" controls={true} title="100e79c0-a79d-4096-ba55-e635eca649dc.wav" class="mdx-audio"/>

#### Flexible audio tag control capabilities

{/* feishu-style:text-align:left */}
In addition to paragraph-level natural language instructions, the model also supports inline audio tags, which are used to precisely control emotions, states, or styles at specific positions in the text. The tags support both Chinese and English languages and open text descriptions, allowing flexible mixing within the same paragraph of text. From simple emotional annotations to complex arrangements with multi-tag overlay and fine-grained layout, the model can express stably, demonstrating excellent performance in both the expressiveness of tags and the stability of combinations. 

{/* feishu-style:text-align:left */}
Text：

{/* feishu-style:text-align:left */}
(调侃) 老张你当时不是说这条航线稳得很吗……

{/* feishu-style:text-align:left */}
(模仿自信，提高音量) “系统全绿，放心走。”

{/* feishu-style:text-align:left */}
(突然停顿) ……现在呢？

{/* feishu-style:text-align:left */}
(爆发，愤怒压不住) 现在整艘船都在报警！你管这叫“放心”？！

{/* feishu-style:text-align:left */}
(声音变轻) 不过……你看那外面，裂开的星云像在呼吸一样。

{/* feishu-style:text-align:left */}
(急促｜呼喊) 别断通讯！喂！再撑十秒！十秒！！

{/* feishu-style:text-align:left */}
(低声｜情绪塌陷般平静) ……算了。

{/* feishu-style:text-align:left */}
(轻笑｜带点释然) 也挺好，至少是一起看的。

{/* feishu-style:text-align:left */}
Audio：

<mimo-audio src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/audios/VwqEbtceyopW54xkuMhcTjE0nIf.mp3" controls={true} title="星云.wav" class="mdx-audio"/>

#### Rich text comprehension ability

{/* feishu-style:text-align:left */}
Even without any prompt or label - just a plain text - the model can directly convey the rhythm and emotion within it. The pauses of punctuation and the undulations of sentence structure will be naturally presented; the emotional arcs hidden in the text, from calm narration to intense twists, can be actively captured by the model; even the speaker's identity (age, temperament, character type) revealed between the lines will automatically be reflected in the voice. In other words: the simplest plain text, when given to it, can still return a vivid and lifelike performance. 

{/* feishu-style:text-align:left */}
Text：

{/* feishu-style:text-align:left */}
Ten... nine... eight... seven... six... five... four... three... TWO... ONE... ZERO! LAUNCH! LAUNCH! WE HAVE LIFTOFF! GO GO GO! SHE'S CLIMBING! ALTITUDE 1,000... 5,000... 10,000 FEET AND CLIMBING! BEAUTIFUL! AB-SO-LUTE-LY BEAUTIFUL!

{/* feishu-style:text-align:left */}
Audio：

<mimo-audio src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/audios/SeyubvO3So5lUIxXnKfcH9vSn5e.mp3" controls={true} title="climb_milo.wav" class="mdx-audio"/>

### Model Series

#### mimo-v2.5-tts

{/* feishu-style:text-align:left */}
 It comes with multiple high-quality voices, covering a variety of usage scenarios. Each voice has been professionally tuned, with natural pronunciation and emotional resonance, allowing you to enjoy high-quality speech synthesis right out of the box.  **Welcome everyone to visit Xiaomi MiMo Studio for voice previews:**   

{/* feishu-style:text-align:left */}
https://aistudio.xiaomimimo.com/#/c

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YJ8zb7ltOouz1wxKmr0cRXSXnCg.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YJ8zb7ltOouz1wxKmr0cRXSXnCg.jpeg" width="364" height="318" alt="图片" style="margin: 16px auto;" />

#### mimo-v2.5-tts-voicedesign

{/* feishu-style:text-align:left */}
The timbre design is aimed at scenarios where " **I have a voice in my heart, but the world doesn't have one yet** ": game NPCs, animated characters, virtual LIVE creators, brand IPs, atypical voices of audio dramas - these are difficult to choose directly from the timbre library and are not suitable for human cloning.

{/* feishu-style:text-align:left */}
This model supports **generating a brand new timbre from scratch through natural language descriptions** ,  without the need for any reference audio. Users can freely use any descriptive dimensions such as age, gender, accent, timbre, vocalization style, personality, etc. - for example, "an elderly Eastern European scholar, with a deep, slightly hoarse voice and a slow speaking rhythm" or "a vibrant young girl, with a clear voice and a slight upward inflection at the end of sentences" - and the model can synthesize the corresponding character timbre. 

{/* feishu-style:text-align:left */}
Thanks to large-scale pre-training, the model can also reasonably interpret **complex, ambiguous, or even contradictory** descriptions, rather than being limited to coarse-grained labels such as "male/female/young/old". This enables timbre design not only to generate unique voices that are difficult for real people to provide, but also to accurately reproduce the voice lines of a certain type of character. 

{/* feishu-style:text-align:left */}
**Case1**

{/* feishu-style:text-align:left */}
Instruct :

{/* feishu-style:text-align:left */}
一位中年男性，说标准普通话，嗓音低沉有磁性，带有轻微的沙哑质感，像纪录片旁白解说员，沉稳而有感染力。

{/* feishu-style:text-align:left */}
Text：

{/* feishu-style:text-align:left */}
当最后一缕阳光消失在地平线之下，这片沉睡了亿万年的大地开始显露它真正的面貌。在这寂静的荒野中，每一块岩石都记录着时间的流逝，每一阵风都在诉说着古老的故事。

{/* feishu-style:text-align:left */}
Audio：

<mimo-audio src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/audios/MkBPbmGlVoLF4Pxpf7scokGknAb.mp3" controls={true} title="纪录片.wav" class="mdx-audio"/>

{/* feishu-style:text-align:left */}
**Case2**

{/* feishu-style:text-align:left */}
Instruct :

{/* feishu-style:text-align:left */}
一位年迈的老先生，说带北方口音的普通话，语速缓慢而沉稳，嗓音略带沙哑和沧桑感，仿佛一位饱经风霜的老爷爷在讲故事，充满岁月的智慧。

{/* feishu-style:text-align:left */}
Text：

{/* feishu-style:text-align:left */}
我这辈子啊，走南闯北六十多年。见过最热闹的集市，也见过最安静的戈壁。到头来才明白一个道理——这人哪，不在走了多远的路，在于记住了多少风景。年轻人，别光顾着赶路，偶尔也停下来看看天。

{/* feishu-style:text-align:left */}
Audio：

<mimo-audio src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/audios/EHVKbKMaMor5CnxAl4Lc4lTVnmc.mp3" controls={true} title="老年.wav" class="mdx-audio"/>

#### mimo-v2.5-tts-voiceclone

{/* feishu-style:text-align:left */}
Voice cloning is used to enable the model **to speak in the voice you specify**—replicating a real-life podcaster, voice actor, brand spokesperson, or the user themselves.

{/* feishu-style:text-align:left */}
**Simply provide a reference audio as short as a few seconds** , and without any additional training, annotation, or fine-tuning process, the model can directly reproduce the speaker's timbre and be immediately available. The reproduced voice not only retains the timbre identity of the original speaker but also preserves personal characteristics such as breath, rhythm, and habitual pauses. 

{/* feishu-style:text-align:left */}
The cloned timbre can **reuse all the control capabilities of this series of models** — natural language instructions, audio tags, and director-level scripts can all continue to be used in combination. The reproduced voice not only "sounds like the original person" but can also perform according to the style and emotion you specify. 

{/* feishu-style:text-align:left */}
Prompt：

<mimo-audio src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/audios/EzmLbRLwjocZSOxX2PVcqKezn5g.mp3" controls={true} title="2.wav" class="mdx-audio"/>

{/* feishu-style:text-align:left */}
Instruct：

{/* feishu-style:text-align:left */}
用尖锐刻薄的嗓音，带着狐假虎威的得意感说话，在提到大人物的身份时故意放慢语速并加重语气，营造压迫感。

{/* feishu-style:text-align:left */}
Text：

{/* feishu-style:text-align:left */}
你以为我是谁，也敢在这儿跟我耍横？我告诉你，站在我身后的那个人，说出来吓死你——是当今的——万岁爷！你今天要是不给我个说法，我让你这铺子明天就开不了门。

{/* feishu-style:text-align:left */}
Audio：

<mimo-audio src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/audios/YfVPb3dikonP11xfHK3cqEo9nEd.mp3" controls={true} title="3_audio.wav" class="mdx-audio"/>

## mimo-v2.5-asr: Understand every expression of yours, no matter how complex

{/* feishu-style:text-align:left */}
If TTS enables voice to become a creative tool at the "output" end, then ASR opens the door to all this at the "input" end. In real-world scenarios, being able to clearly and accurately understand speech amidst language switching, background noise, and speakers with strong dialect accents is what truly makes a good speech recognition system. 

{/* feishu-style:text-align:left */}
mimo-v2.5-asr, as the auditory foundation of the whole-link speech model series, has achieved industry-leading levels in complex real-world scenarios such as Chinese-English bilingual, Chinese dialects, Code-Switch, strong noise, multi-speaker, and high knowledge density. It is not just about converting clear speech into text, but also enabling agents to capture every word and phrase worthy of understanding in noisy real-world sounds.

### Core Features

{/* feishu-style:text-align:left */}
Chinese dialects: Supports dialects such as Wu dialect, Cantonese, Minnan dialect, Sichuan dialect, etc.

{/* feishu-style:text-align:left */}
Complex English Scenarios: Achieved leading performance on the Open ASR Leaderboard in complex English scenarios such as AMI

{/* feishu-style:text-align:left */}
Code-Switch: Free and smooth speech transcription for Chinese-English Code-Switch, no need to pre-set language labels

{/* feishu-style:text-align:left */}
Song Recognition: Recognizes lyrics of Chinese and English songs, maintaining high accuracy in scenarios where accompaniment and vocals are mixed 

{/* feishu-style:text-align:left */}
Strong noise scenario: Maintains robust recognition in complex acoustic environments such as high noise and far-field sound pickup

{/* feishu-style:text-align:left */}
Multi-speaker: Supports accurate transcription of multi-person cross-dialogue scenarios, such as meeting scenarios

{/* feishu-style:text-align:left */}
Strong Knowledge Association: Precise identification of knowledge-intensive content such as ancient poems, technical terms, personal names, and place names

{/* feishu-style:text-align:left */}
Native Punctuation: Outputs punctuation natively by combining speech prosody and semantics, with the transcription results ready for immediate use without post-processing

### Performance

{/* feishu-style:text-align:left */}
mimo-v2.5-asr has achieved the current optimal or highly competitive results across multiple dimensions, including general Chinese and English, Chinese dialects, Code-Switch, and lyrics recognition, demonstrating its stable advantages across scenarios and languages. The following are representative evaluation results:

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TdZBb6yCIo2L0bxH23IcYuJgnob.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TdZBb6yCIo2L0bxH23IcYuJgnob.png" width="614" height="436" alt="图片" style="margin: 16px 0;" />

{/* feishu-style:text-align:left */}
For Agent applications, content creation tools, conferencing systems, and voice interaction products, this is a truly verified auditory foundation in complex real-world speech. 

## How to Use 

### MiMo-V2.5-TTS Series

{/* feishu-style:text-align:left */}
To assist developers in exploring more scenarios,**mimo-v2.5-tts, mimo-v2.5-tts-voicedesign, and mimo-v2.5-tts-voiceclone** are all available for free on the **Xiaomi MiMo API** Open Platform **for a limited time:** 
https://platform.xiaomimimo.com/docs/usage-guide/speech-synthesis-v2.5

{/* feishu-style:text-align:left */}
Meanwhile, everyone is welcome to visit **Xiaomi MiMo Studio** for a quick experience:https://aistudio.xiaomimimo.com/#/c

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AnrTbxLbJoGF7Pxs0s2cMOO3neg.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AnrTbxLbJoGF7Pxs0s2cMOO3neg.jpeg" width="3022" height="1716" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
For more cases, please refer to [https://mimo.xiaomi.com/mimo-v2-5-tts](https://mimo.xiaomi.com/mimo-v2-5-tts)

### mimo-v2.5-asr

{/* feishu-style:text-align:left */}
**mimo-v2.5-asr has now open-sourced its model weights and code**, enabling developers and researchers to directly use or conduct secondary development.

> Demo page: [https://mimo.xiaomi.com/mimo-v2-5-asr](https://mimo.xiaomi.com/mimo-v2-5-asr)
>
> Project Open Source Address: [https://github.com/XiaomiMiMo/MiMo-V2.5-ASR](https://github.com/XiaomiMiMo/MiMo-V2.5-ASR)
>
> Weight Open Source Address: https://huggingface.co/XiaomiMiMo/MiMo-V2.5-ASR
>
> Huggingface space: https://huggingface.co/spaces/XiaomiMiMo/MiMo-V2.5-ASR

## Agent Tool Call Support 

{/* feishu-style:text-align:left */}
To facilitate everyone's quick integration of speech capabilities into Agent applications, we have fully open-sourced the access Skill for mimo-v2.5-tts related models. Welcome to visit the repository to pull and use: 

{/* feishu-style:text-align:left */}
[https://github.com/XiaomiMiMo/MiMo-Skills](https://github.com/XiaomiMiMo/MiMo-Skills)

### Sound is just the starting point

{/* feishu-style:text-align:left */}
Beyond the MiMo-V2.5-TTS Series, we would like to answer a question:

{/* feishu-style:text-align:left */}
What will audio creation look like when mimo-v2.5-tts understands "expression", mimo-v2.5-pro understands "planning", and mimo-v2.5 understands "listening"? 

{/* feishu-style:text-align:left */}
**The answer is: a complete, closed-loop Agent-style creative chain.** 

- mimo-v2.5-pro —— Planning and screenwriting, breaking down tasks, writing scripts, arranging rhythm, and determining the editing sequence.

- MiMo-V2.5-TTS Series —— Timbre and Creatives, Voice Design generates timbre, Voice Clone synthesizes content.

- mimo-v2.5 —— Listening back and evaluation, checking if the character is consistent, if the rhythm is correct, and if it deviates from the user's original intention.

{/* feishu-style:text-align:left */}
An example:

> Create a scene of a summer afternoon lasting about 2 minutes. Grandpa (in his 70s, with a Beijing hutong accent, hoarse voice, drawn-out speech, lowered voice when concentrating on chess, and a booming laugh with a table slap) is playing chess under a pagoda tree. A 5-year-old grandson is squatting beside, watching ants, and occasionally interrupting with childish questions (clear, with rising intonation at the end, higher when excited, and occasional unclear pronunciation). Grandpa's tone is solemn when he gets serious, but immediately softens into a laughing scold when interrupted by his grandson.

{/* feishu-style:text-align:left */}
Users only provide a single sentence, and the finished product is generated automatically:

<video src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/videos/BZdlbxZxGotw9GxYkf6cjjhLnob.mp4" controls={true} playsInline={true} title="video-60s.mp4" class="mdx-video" />

<mimo-audio src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/audios/Nv6AbNOe6oGrIFxEnc2cQaORnzg.mp3" controls={true} title="agent-wav.wav" class="mdx-audio"/>

{/* feishu-style:text-align:left */}
**Some may say it's a threshold, but being able to listen, think, and collaborate is what truly matters.** 

## Next step

1. **Larger-scale speech pre-training and post-training with reinforcement learning:**  MiMo-V2.5-TTS-Series demonstrates the significant benefits of large-scale pre-training and post-training, expanding the scale of both: through more data, larger models, and stronger computing power, enabling more powerful speech intelligence to emerge from scale; more refined reward modeling and reinforcement learning algorithms drive the model towards higher-order speech expression intelligence. 

1. **Universal Audio Generation:** Speech is just the first step. We are expanding our capabilities to more generalized audio generation: environmental sound effects, action sounds, ambient backgrounds, and even short musical phrases and melody segments—gradually modeling a complete sonic world. We believe that a true universal audio model is not simply piecing together speech, sound effects, and music, but enabling them to understand each other and collaborate in the same space.

1. **Contextual understanding ability:** Speech expression has never been an isolated sentence game. The reason people can "read correctly" is because they understand the context—knowing what happened before and understanding where the current sentence fits into the overall narrative. Contextual understanding means that the model is no longer just a "tool for executing sentences one by one," but an expresser who understands the context of the story. This is a crucial step towards truly general speech intelligence.

1. **General Speech Understanding Ability:** Our goal is to ensure that "real-world norms" such as dialects, noise, and Chinese-English mixtures no longer become the weak points of speech recognition. In the future, we will continue to expand the coverage of more dialects and deepen context awareness capabilities, enabling speech recognition to evolve from "transcription" to "understanding".

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/J1qQbcRZmoXwglxvlSccY2ptnQb.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/J1qQbcRZmoXwglxvlSccY2ptnQb.jpeg" width="618" height="225" alt="图片" style="margin: 16px auto;" />


--- DOCUMENT: Xiaomi MiMo-V2.5 Series Large Model Launches Public Beta ---
URL: https://mimo.mi.com/static/docs/news/latest/v2.5-news.md

# Xiaomi MiMo-V2.5 Series Large Model Launches Public Beta

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/T8xCbyh88oQkeFx4KtPcVWpMn0c.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/T8xCbyh88oQkeFx4KtPcVWpMn0c.jpeg" width="557" height="557" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
Today, the Xiaomi MiMo-V2.5 series of models officially launched its public beta.

{/* feishu-style:text-align:left */}
The Xiaomi MiMo-V2.5 series includes mimo-v2.5, v2.5-pro, V2.5-TTS Series, and v2.5-asr. 

{/* feishu-style:text-align:left */}
Stronger reasoning, more stable agents, longer context, stronger instruction following and understanding of ambiguous instructions, better full-modal perception and understanding —— this is a comprehensive leap from "usable" to "user-friendly".

{/* feishu-style:text-align:left */}
Meanwhile, we have also optimized the Token Plan pricing plan —— making the world's top-notch models easily accessible.

##  mimo-v2.5-pro: Stronger Agent, Longer Focus

{/* feishu-style:text-align:left */}
mimo-v2.5-pro is our most powerful model to date. In dimensions such as **general agent capabilities, complex software engineering, and long-range tasks**, it can already compete head-on with the world's top Agent models (Claude Opus 4.6, GPT-5.4), achieving a comprehensive leap compared to the previous generation mimo-v2-pro.

{/* feishu-style:text-align:left */}
During internal testing, the intelligence level demonstrated by mimo-v2.5-pro has made us rethink the way humans and models collaborate: when paired with a suitable operating framework, it can stably complete long-range tasks involving nearly a thousand rounds of tool calls in a single instance, and its instruction-following ability in the agent scenario has also significantly improved - it can accurately capture implicit requirements in the context and maintain logical consistency over an extremely long period. By now, mimo-v2.5-pro can already undertake truly serious professional work with a higher confidence level. 

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Jsu9bEinQoAo1ZxbgJNc9VV3nDc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/Jsu9bEinQoAo1ZxbgJNc9VV3nDc.png" width="1080" height="722" alt="图片" style="margin: 16px auto;" />

#### Designed for more complex tasks

{/* feishu-style:text-align:left */}
mimo-v2.5-pro is designed for more challenging and complex task objectives. We assign tasks that would take human experts days or even weeks to complete to it, allowing it to independently complete long-term processes while still maintaining extremely high quality. The following are the results it has delivered:

##### **Implement a complete SysY compiler in Rust**

{/* feishu-style:text-align:left */}
This task originated from the Compilation Principles course project at Peking University, requiring the model to implement a complete SysY compiler from scratch in Rust, including a lexical analyzer, a syntax analyzer, AST, Koopa IR code generation, RISC-V assembly backend, and performance optimization. For reference,  **undergraduate students at Peking University usually take** **several weeks to complete this project, while mimo-v2.5-pro only took** **4.3 hours**, completed all tasks after 672 tool calls, and achieved **a perfect score of 233/233 on the hidden test set, demonstrating extremely high productivity value.**  

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ZS0NbihSUoqRRtxffeOckXzCnue.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ZS0NbihSUoqRRtxffeOckXzCnue.jpeg" width="1080" height="726" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
Instead of getting stuck in brute-force trial-and-error, it builds the entire compiler layer by layer: first constructing the complete pipeline framework, then tackling each layer one by one - Koopa IR achieved a perfect score (110/110), the RISC-V backend achieved a perfect score (103/103), and Performance optimization achieved a perfect score (20/20). The first compilation passed **137/233** , with a cold start pass rate of 59%, which means that the architecture was already correct before running any tests. In the 512th round, a refactoring caused lv9/riscv to regress by two test points; the model self-diagnosed, recovered, and continued to progress. 

{/* feishu-style:text-align:left */}
**The long-range task rewards precisely this structured and self-correcting work discipline.** 

##### **Develop a video editor**

{/* feishu-style:text-align:left */}
With just a few simple instructions - "Build a video editor web application" - mimo-v2.5-pro delivered a runnable web application: featuring multi-track timeline, clip trimming, cross-fading, audio mixing, and export processes. The final built codebase amounts to 8,192 lines, involves 1,868 tool invocations, and was completed in 11.5 hours of autonomous work.

##  mimo-v2.5: Overstepping Full-Modal Agent, Million Contexts

{/* feishu-style:text-align:left */}
mimo-v2.5 is a native full-modal large model designed for Agent scenarios, capable of seeing, hearing, and reading simultaneously, and translating understanding into action.

{/* feishu-style:text-align:left */}
This time, mimo-v2.5 brings a key upgrade: 

{/* feishu-style:text-align:left */}
**Agent capabilities comprehensively surpass mimo-v2-pro**

{/* feishu-style:text-align:left */}
In authoritative Agent evaluations such as Claw-Eval, mimo-v2.5 surpasses the level of mimo-v2-pro, is capable of handling daily simple tasks, and at the same time reduces API costs by approximately 50%. 

{/* feishu-style:text-align:left */}
**MultiModal Machine Learning perception comprehensively surpasses mimo-v2-omni**

{/* feishu-style:text-align:left */}
Capabilities such as cross-modal reasoning, video understanding, and chart analysis have been enhanced, approaching and even surpassing industry-leading closed-source models in evaluations such as VideoMME, CharXiv, and MMMU-Pro.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DLIrb2z64odTsCxpfzocBH8tngd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DLIrb2z64odTsCxpfzocBH8tngd.png" width="1080" height="396" alt="图片" style="margin: 16px auto;" />

##  MiMo-V2.5 Full Series: Higher Token Efficiency

{/* feishu-style:text-align:left */}
The entire MiMo-V2.5 series is optimized for Token efficiency, doing more with fewer Tokens. 

{/* feishu-style:text-align:left */}
When achieving the same score on the Agent benchmark list ClawEval: 

- mimo-v2.5-pro saves 42% Token compared to Kimi K2.6

- mimo-v2.5 saves 50% of tokens compared to Muse Spark

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/I1BnbUiTKoQl9QxPoYxczCPEnOf.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/I1BnbUiTKoQl9QxPoYxczCPEnOf.jpeg" width="1080" height="748" alt="图片" style="margin: 16px auto;" />

##  MiMo-V2.5 Full Series: How to Use Them in Combination? 

- mimo-v2.5-pro is specifically designed for long and complex Agent tasks, while mimo-v2.5 covers most general Agent scenarios

- mimo-v2.5 supports native full-modal Agent capabilities, covering images, audio, and video

- mimo-v2.5 has a higher average inference speed and can respond more quickly to latency-sensitive tasks

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AbE5bpAYaovgIlxz2jPcziYznGf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/AbE5bpAYaovgIlxz2jPcziYznGf.png" width="739" height="406" alt="图片" style="margin: 16px auto;" />

##   Token Plan Upgraded and Refreshed 

{/* feishu-style:text-align:left */}
We have made several substantial optimizations suitable for you regarding the Token Plan: 

{/* feishu-style:text-align:left */}
**Credits rate updated, more favorable**

- mimo-v2.5：1x（use 1 Token = 1 Credit）

- mimo-v2.5-pro： 2x（use 1 Token = 2 Credits）

{/* feishu-style:text-align:left */}
**Cancel the billing method of 1 Token = 4 Credits. From now on, the Token Plan will no longer distinguish the Credit multiplier for 256k and 1M context windows.** 

{/* feishu-style:text-align:left */}
**Exclusive Nighttime Discount Rate**

{/* feishu-style:text-align:left */}
From 00:00 to 08:00 Beijing Time every day, the consumption rate of Credits for all models**will be further discounted by 20% on top of the original rate**. 

{/* feishu-style:text-align:left */}
**Enjoy discounts with auto-renewal** 

{/* feishu-style:text-align:left */}
A new "Continuous Monthly Subscription" model has been added. Existing users who activate auto-renewal will enjoy a 30% discount on the next month's subscription, while new users will enjoy a 23% discount on the next month's subscription, both limited to one time. 

{/* feishu-style:text-align:left */}
A new "Annual" subscription cycle has been added. Subscribing once will enjoy an 12% discount for the whole year, and no longer be combined with the first purchase/auto-renewal discount. 

##  Online Benefit: Token Plan users' Credits will be fully reset

{/* feishu-style:text-align:left */}
All users who have purchased the Token Plan (as of 22:00 on April 22, Beijing Time) **will have their Credits quota fully reset to zero**, and the calculation will start anew.

{/* feishu-style:text-align:left */}
Xiaomi MiMo helps you start from scratch and unleash your creativity to the fullest!

> Note: This online welfare only resets the Credits limit, does not reset the package timing, and the validity period of purchased packages remains unchanged.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BubdbNjLlogILoxgG1vca3mxnEf.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/BubdbNjLlogILoxgG1vca3mxnEf.jpeg" width="499" height="523" alt="图片" style="margin: 16px auto;" />

##   is about to be open-sourced 

{/* feishu-style:text-align:left */}
mimo-v2.5-pro and mimo-v2.5 models are about to be globally open-sourced. Stay tuned!


--- DOCUMENT: Xiaomi MiMo is now integrated with the top-tier Agent framework Hermes Agent and offers a two-week free trial ---
URL: https://mimo.mi.com/static/docs/news/latest/hermes-free.md

# Xiaomi MiMo is now integrated with the top-tier Agent framework Hermes Agent and offers a two-week free trial

{/* feishu-style:text-align:left */}
The Xiaomi MiMo-V2 series now officially supports Hermes Agent! 

{/* feishu-style:text-align:left */}
As the flagship base for the Agent era, the Xiaomi MiMo-V2 series of large models has officially joined hands with the world's leading Agent open-source framework Hermes Agent to achieve official integrated access. 

{/* feishu-style:text-align:left */}
**Hermes Agent is：**  

- One of the most globally watched open-source Agent frameworks currently

- It has the capabilities of self-evolution and cross-session memory, automatically accumulates experience from tasks, and becomes stronger with more use 

- Supports cross-platform communication

{/* feishu-style:text-align:left */}
mimo-v2-pro, with its 1M long context capability, native strong tool invocation, and in-depth Agent-specific optimization, is fully compatible with core features of Hermes Agent such as self-evolution skills, cross-session memory, and complex workflows.

{/* feishu-style:text-align:left */}
mimo-v2-omni further expands the boundaries of perception, integrating the full-modal understanding capabilities of images, videos, audio, and text, enabling Hermes Agent to become a true full-modal agent that can see, understand, and act.

{/* feishu-style:text-align:left */}
**We sincerely invite developers worldwide to try it out, with a two-week free trial** 

- **Free trial period:**   April 8 - April 22, 12:00 (Beijing Time, UTC+8), a total of two weeks. 

- **Usage:**   Update Hermes Agent to the latest version, and you can call Xiaomi mimo-v2-pro, omni, and flash models for free via Nous Portal.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TFqGbro9kobTwnx3mwjc5A7PnNe.webp" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/TFqGbro9kobTwnx3mwjc5A7PnNe.webp" width="798" height="392" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
From "task execution" to "self-evolution", MiMo, in collaboration with Hermes Agent, enables your AI Agent to become "smarter with use".


--- DOCUMENT: Xiaomi MiMo Token Plan Brand New Release ---
URL: https://mimo.mi.com/static/docs/news/latest/token-plan-release.md

# Xiaomi MiMo Token Plan Brand New Release

{/* feishu-style:text-align:left */}
Since 2025, the capabilities of large models have been continuously redefined. However, for most developers and users, "affordability" remains a more fundamental issue than "usability". Under the pay-as-you-go model, every invocation is accompanied by uncertainty about costs.

{/* feishu-style:text-align:left */}
We don't want it to be this way. We believe that  **good technology should not be a privilege reserved for only a few; it should, like water and electricity, become a readily accessible productivity tool for everyone at a predictable price.** 

{/* feishu-style:text-align:left */}
Therefore, we officially launch the Xiaomi MiMo Token Plan.

### MiMo Token Plan: Simplify Complexity

{/* feishu-style:text-align:left */}
Our original intention in designing the Token Plan was to ensure that the billing method is transparent, simple, and straightforward enough for any user to understand and use with ease.

{/* feishu-style:text-align:left */}
**1. Transparent design, simple and straightforward** - Unified Credit point system, converting credit consumption based on token usage, helping you easily plan your usage.

> - mimo-v2-omni 256k Context: 1x (1 Token Consumed = 1 Credit)
>
> - mimo-v2-pro 256k Context: 2x (1 Token Consumed = 2 Credits)
>
> - mimo-v2-pro 256k~1M Context: 4x (1 Token Consumed = 4 Credits)
>
> - mimo-v2-tts: 0x (Limited-time free, no Credit consumption)

{/* feishu-style:text-align:left */}
**2. No 5-hour token usage limit** —— Supports concentrated token consumption, enabling high-intensity lobster farming or programming with a full experience and no interruptions.

{/* feishu-style:text-align:left */}
**3. Users who purchase the package can enjoy the priority internal testing experience right for the new model**—advanced and user-friendly, one step ahead.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/C4NKbWi3NoXjfzxBCuscOiyinFe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/raw/C4NKbWi3NoXjfzxBCuscOiyinFe.png" width="466" height="478" alt="图片" style="margin: 16px auto;" />

### Four-tier pricing, designed for you

{/* feishu-style:text-align:left */}
Token Plan offers four tiers of packages, so no matter your frequency and depth of AI usage, you can find a suitable plan:

- **Lite (China: ¥39/month. Overseas: $6/month)**  —— 60M Credits, can execute approximately **120 medium to complex tasks**. Suitable for explorers new to AI development, starting at the price of a cup of coffee.

- **Standard (China: ¥99/month. Overseas: $16/month)**  —— 200M Credits, capable of performing approximately **400 medium to complex tasks**. A primary solution designed for work and developer users who rely on AI for daily efficiency improvement.

- **Pro (China: ¥329/month. Overseas: $50/month)**  —— 700 million (700M) Credits, capable of performing approximately **1,400 medium to complex tasks**. Designed for professional users who deeply integrate AI into their workflows.

- **Max (China: ¥659/month. Overseas: $100/month)**  —— 1.6 billion (1600M) Credits, capable of executing approximately **3200 medium to complex tasks**. Designed for developers with all-day, high-intensity usage, offering an almost unrestricted usage experience.

> All packages enjoy a **12% discount** on the first purchase, and this discount is limited to 1 time only.
>
> <img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/LgVzbjs0Wovbkqxj7ImcjZnVnpg.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/raw/LgVzbjs0Wovbkqxj7ImcjZnVnpg.png" width="432" height="626" alt="图片" style="margin: 16px auto;" />

### Specifically adapted for mainstream AI tools 

{/* feishu-style:text-align:left */}
Specifically designed for mainstream AI tools and development platforms such as Claude Code, OpenClaw, OpenCode, Kilo Code, Cline, etc., to help you efficiently boost productivity. 

### This is just the beginning 

{/* feishu-style:text-align:left */}
Auto-renewal, plan upgrades, and more flexible usage management are all under development. If you have any ideas or suggestions during use, we sincerely look forward to your feedback.

{/* feishu-style:text-align:left */}
Token Plan is the new starting point for MiMo, and our goal has never changed: **to create the best models, set the most reasonable prices, and enable more people to truly use them.** 

{/* feishu-style:text-align:left */}
👉 Purchase Now: [Xiaomi MiMo Open Platform](https://platform.xiaomimimo.com/)


--- DOCUMENT: Xiaomi MiMo Agent Framework Call Free Trial Extension for One Week ---
URL: https://mimo.mi.com/static/docs/news/latest/free-trial-extension.md

# Xiaomi MiMo Agent Framework Call Free Trial Extension for One Week

{/* feishu-style:text-align:left */}
Since the global release of the new models in the Xiaomi MiMo-V2 series on March 19, 2026, the mimo-v2-pro/omni has been enthusiastically pursued and widely concerned by developers worldwide, especially the flagship model mimo-v2-pro in the global call volume ranking of OpenRouter**has continuously ranked No. 1 in the daily, weekly, and trending lists**. 

{/* feishu-style:text-align:left */}
In addition, our joint operation activities carried out together with  **top Agent frameworks such as OpenClaw, OpenCode, KiloCode, Cline, and BlackBoxAI**  are also highly popular among users. 

{/* feishu-style:text-align:left */}
Therefore, we have decided - **to extend the "XiaomiMiMo Launches First Week Free Trial in Collaboration with Global Top Agent Framework" event from the originally scheduled one-week free trial to two weeks,**  and the free trial period will be extended to:  **12:00 PM, April 2, 2026, Beijing Time (GMT+8).** 

{/* feishu-style:text-align:left */}
For the limited-time free access methods of each platform, please refer to:[Xiaomi MiMo Partners with Top Agent Framework : First Week Free](https://platform.xiaomimimo.com/#/docs/news/first-week-free)

{/* feishu-style:text-align:left */}
AI without barriers, innovation without limits. We sincerely invite global developers to fully unleash the powerful productivity of the combination of Xiaomi MiMo large model and top-tier Agent framework.


--- DOCUMENT: Xiaomi MiMo Partners with Top Agent Framework : First Week Free ---
URL: https://mimo.mi.com/static/docs/news/previous-news/first-week-free.md

# Xiaomi MiMo Partners with Top Agent Framework : First Week Free

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/RgAbbHPPQoTpLOxnirNcPr1EnXr.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/RgAbbHPPQoTpLOxnirNcPr1EnXr.png" width="803" height="355" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
mimo-v2-pro, mimo-v2-omni, and mimo-v2-tts are now available. To meet global developers' anticipation for Xiaomi MiMo's new base models, Xiaomi MiMo has partnered with five agent frameworks — OpenClaw, OpenCode, KiloCode, Cline, and BLACKBOXAI — offering free API access worldwide for one week.

{/* feishu-style:text-align:left */}
**Note: mimo-v2-pro and mimo-v2-omni can only be used for free in the above five agent frameworks, see below for detailed instructions. To call the Model API directly, see** [**Pricing and Rate Limits**](https://platform.xiaomimimo.com/#/docs/pricing) **for pricing standards.** 

## 01 / OpenClaw

{/* feishu-style:text-align:left */}
**Free for a Limited Time** — A Highly Anticipated General-Purpose Agent Framework. AI That Truly Gets Things Done.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DrWSbyYP9oqCBAxuR2OcO2yZnGd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DrWSbyYP9oqCBAxuR2OcO2yZnGd.png" width="692" height="474" alt="图片" style="margin: 16px auto;" />

#### Integration

{/* feishu-style:text-align:left */}
Get an API Key from OpenRouter and configure it in your OpenClaw:

- In chat: `/model openrouter/xiaomi/mimo-v2-pro` (or v2-omni)

- In terminal: `openclaw models set openrouter/xiaomi/mimo-v2-pro` (or v2-omni)

- Or edit config: set `model.primary` to `openrouter/xiaomi/mimo-v2-pro` (or v2-omni)

{/* feishu-style:text-align:left */}
See details on [OpenClaw provides free access to MiMo-V2-Pro/MiMo-V2-Omni replicas via Openrouter](https://platform.xiaomimimo.com/#/docs/integration/openclaw-with-openrouter) .

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/HfFhbA7DaoyE2SxbmbPcVgxynld.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/HfFhbA7DaoyE2SxbmbPcVgxynld.png" width="691" height="625" alt="图片" style="margin: 16px auto;" />

## 02 / OpenCode

{/* feishu-style:text-align:left */}
**Free for a Limited Time** — Open-Source AI Coding Agent. 120K+ GitHub Stars. 5M+ Developers Monthly.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/JE0ZbgLiFoMqcHxYXPscSzeVnHc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/JE0ZbgLiFoMqcHxYXPscSzeVnHc.png" width="698" height="262" alt="图片" style="margin: 16px auto;" />

#### Integration

{/* feishu-style:text-align:left */}
From the terminal, desktop app, or IDE extension, select MiMo V2 Pro / MiMo V2 Omni (FREE tag) under OpenCode → Zen.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/HRXbbkfF6o7JBBxT76icaD7Zn0u.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/HRXbbkfF6o7JBBxT76icaD7Zn0u.jpeg" width="700" height="427" alt="图片" style="margin: 16px auto;" />

## 03 / KiloCode

{/* feishu-style:text-align:left */}
**Free for a Limited Time —** A Full-Featured AI Engineering Platform for Developers. 1M+ Kilo Developers.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OaOmb9EotoM6eoxKrDrcXIu5nfI.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/OaOmb9EotoM6eoxKrDrcXIu5nfI.png" width="705" height="494" alt="图片" style="margin: 16px auto;" />

#### Integration

{/* feishu-style:text-align:left */}
From the terminal or IDE extension, select MiMo V2 Pro / MiMo V2 Omni (FREE tag) under Kilo Gateway.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/EBAAbV8kNo29Btx6doRcKZltnpb.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/EBAAbV8kNo29Btx6doRcKZltnpb.jpeg" width="710" height="345" alt="图片" style="margin: 16px auto;" />

## 04 / Cline

{/* feishu-style:text-align:left */}
**Free for a Limited Time —** AI Coding Assistant That Helps Developers Build and Refactor High-Quality Software at 2x Speed.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FEv0bcYLzoL83exEPkac3bPKngb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FEv0bcYLzoL83exEPkac3bPKngb.png" width="715" height="384" alt="图片" style="margin: 16px auto;" />

#### Integration

{/* feishu-style:text-align:left */}
From the terminal or IDE extension, select Cline as the Provider and choose mimo-v2-pro (FREE tag).

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ToQEbkzrOoAyP8xst0AclL4GnTo.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/ToQEbkzrOoAyP8xst0AclL4GnTo.jpeg" width="719" height="549" alt="图片" style="margin: 16px auto;" />

## 05 / BLACKBOX.AI

{/* feishu-style:text-align:left */}
One of the Fastest-Growing Coding Agents Globally — Committed to Redefining How You Write Code with AI.

{/* feishu-style:text-align:left */}
Blackbox.AI is Currently in Final Testing and Coming Soon. Stay Tuned to Blackbox.AI's Official X Posts.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/JpcYbeWS7owoE8xZHeMcIvFdn0g.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/JpcYbeWS7owoE8xZHeMcIvFdn0g.png" width="723" height="356" alt="图片" style="margin: 16px auto;" />

#### Integration

{/* feishu-style:text-align:left */}
Select the model in the terminal, desktop app, or IDE extension to integrate.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/JNzqbHvpyoiRQ8xhTCRc5oZjnUC.jpeg" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/JNzqbHvpyoiRQ8xhTCRc5oZjnUC.jpeg" width="728" height="389" alt="图片" style="margin: 16px auto;" />

## 06 / mimo-v2-tts

{/* feishu-style:text-align:left */}
**Free for an Extended Period.**  — Xiaomi's In-House Text-to-Speech Foundation Model. Empowering Agents with Warm, Expressive, and Soulful Voice.

#### Integration

{/* feishu-style:text-align:left */}
Access via the official API platform.

{/* feishu-style:text-align:left */}
For detailed integration, refer to: [MiMo-V2-TTS Usage Guide](https://platform.xiaomimimo.com/#/docs/usage-guide/speech-synthesis)

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/QdYQbLmfZofJtNxMQ16ca62Lnaf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/QdYQbLmfZofJtNxMQ16ca62Lnaf.png" width="520" height="302" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
AI Without Barriers. Innovation Without Limits. Global Developers — Unleash the Power of Trillion-Parameter Models Paired with Top-Tier Agent Frameworks.


--- DOCUMENT: Xiaomi MiMo-V2-Pro: Flagship Foundation Model towards Agent Era ---
URL: https://mimo.mi.com/static/docs/news/previous-news/v2-pro-release.md

# Xiaomi MiMo-V2-Pro: Flagship Foundation Model towards Agent Era

{/* feishu-style:text-align:left */}
Today, we are releasing Xiaomi mimo-v2-pro, Xiaomi’s flagship foundation model for the agent era.

{/* feishu-style:text-align:left */}
Xiaomi mimo-v2-pro is built for demanding real-world Agent workflows. It has over **1T** total parameters, with **42B** active parameters, uses an innovative hybrid attention architecture, and supports an ultra-long context window of up to **1M** tokens. Based on the strong foundation model, we continue to scale compute across a broader range of agent scenarios, further expanding the action space of intelligence and achieving an important generalization leap from Coding to Claw.

{/* feishu-style:text-align:left */}
On the global authoritative model intelligence ranking by Artificial Analysis, mimo-v2-pro ranks eighth worldwide and second in China.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FZw6bk5M6odYsSxODBOcZds6nAb.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/FZw6bk5M6odYsSxODBOcZds6nAb.png" width="4129" height="1141" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
In agent frameworks such as OpenClaw and Claude Code, mimo-v2-pro shows excellent end-to-end task completion ability. It can handle complex workflow orchestration, long-horizon planning, and precise tool use without human intervention, while reliably delivering final results. In overall hands-on experience, it has surpassed Claude Sonnet 4.6 and is approaching Opus 4.6, while its API pricing is only one-fifth of theirs, lowering the barrier to using frontier intelligence.

## A major leap in foundation capabilities

{/* feishu-style:text-align:left */}
By scaling both parameters and compute, mimo-v2-pro reaches to a larger and stronger model foundation.

- **Trillion-parameter scale, efficient architecture**: Total parameters exceed 1T, with 42B active parameters, about 3x larger than the previous mimo-v2-flash. It continues to use the innovative Hybrid Attention mechanism introduced in mimo-v2-flash, with the hybrid ratio further increased from 5:1 to 7:1. This keeps inference efficient even with the large increase in model size, while also supporting 1M-token context. A lightweight MTP (Multi-Token Prediction) layer enables fast generation.

- **From Chat to Agent**: By scaling during post-training across a broader set of Agent tasks, the model is no longer limited to “answering questions” or “generating polished demos.” It is built to complete tasks. We aim to integrate it deeply into productivity scenarios so it can serve as the “brain” behind working systems and continuously deliver results with real-world impact.

- **Real-world experience beyond benchmark rankings**: mimo-v2-pro performs strongly across benchmarks that measure key model capabilities. In Coding Agent, general Agent, and Tool Use tasks, it is in the same tier as Claude 4.5 Sonnet, GPT5.2, and Gemini 3.0 Pro, showing leading intelligence. We remain focused on training and optimization guided by actual user experience, always paying close attention to how the model performs in real applications.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/IAt5buS6To3zKExg1RZcsGhhnPc.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/IAt5buS6To3zKExg1RZcsGhhnPc.png" width="584" height="380" alt="图片" style="margin: 16px auto;" />

## A flagship model built for Agents

{/* feishu-style:text-align:left */}
mimo-v2-pro is deeply optimized specifically for Agent scenarios.

### The native brain for OpenClaw

{/* feishu-style:text-align:left */}
OpenClaw is a general-purpose agent framework that has recently gained strong attention in the open-source community. As the core engine behind frameworks like this, the upper limit of the underlying model directly determines the system’s real-world performance. mimo-v2-pro is trained with SFT and RL on complex and diverse Agent scaffolds, giving it stronger tool-use and multi-step reasoning abilities. 

{/* feishu-style:text-align:left */}
On OpenClaw’s standard benchmark leaderboards, PinchBench and ClawEval, mimo-v2-pro ranks among the best in the world. At the same time, with its 1M-token context window, mimo-v2-pro can comfortably support demanding real-world Claw application flows. Hunter Alpha shown below is an early anonymous version of mimo-v2-pro.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/XDC4bTG01opiD1xNgphcTj9gnih.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/XDC4bTG01opiD1xNgphcTj9gnih.png" width="570" height="541" alt="图片" style="margin: 16px auto;" />

### Continuous Evolution of Coding Capabilities

{/* feishu-style:text-align:left */}
Going beyond mere "Vibe Coding", mimo-v2-pro is capable of participating in more rigorous code engineering construction.

{/* feishu-style:text-align:left */}
In in-depth evaluations by internal engineers at Xiaomi, mimo-v2-pro's user experience has approached that of Claude Opus 4.6, demonstrating advanced code intelligence: it boasts superior system design and task planning capabilities, more elegant coding styles, and more efficient, direct problem-solving pathways.

{/* feishu-style:text-align:left */}
During the "Hunter Alpha" anonymous testing phase, the most frequently called apps were mostly programming-specific tools, which confirm mimo-v2-pro's high usability and reliability in real-world R&D scenarios.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/RxMnbgF97owiMPxcclPcrF5CnJd.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/RxMnbgF97owiMPxcclPcrF5CnJd.png" width="544" height="304" alt="图片" style="margin: 16px auto;" />

## 1M Context Window, Open API

{/* feishu-style:text-align:left */}
The mimo-v2-pro model is now officially available via API with pricing:

- Within 256K: Input at $1 / 1M tokens, Output at $3 / 1M tokens

- 256K ~ 1M: Input at $2 / 1M tokens, Output at $6 / 1M tokens

{/* feishu-style:text-align:left */}
Visit [https://platform.xiaomimimo.com](https://platform.xiaomimimo.com/) to get started.


--- DOCUMENT: Xiaomi MiMo-V2-Omni: Omni-Modal Agentic Foundation Model that Sees, Understands and Acts ---
URL: https://mimo.mi.com/static/docs/news/previous-news/v2-omni-release.md

# Xiaomi MiMo-V2-Omni: Omni-Modal Agentic Foundation Model that Sees, Understands and Acts

{/* feishu-style:text-align:left */}
Today, we are thrilled to announce Xiaomi’s omni‑modal foundation model for agent era: Xiaomi mimo-v2-omni.

{/* feishu-style:text-align:left */}
Designed specifically for complex real‑world multimodal interaction and execution scenarios, mimo-v2-omni is built from the ground up as a unified all‑modal foundation that integrates text, vision, and speech. Its unified architecture deeply binds perception and action, overcoming the traditional limitation of models that prioritize understanding over execution.

{/* feishu-style:text-align:left */}
Natively equipped with multimodal perception, tool invocation, function execution, and GUI operation capabilities, mimo-v2-omni seamlessly integrates with major agent frameworks. It enables a true leap from understanding to control, drastically lowering the barrier to deploying all‑modal agents.

## Perception Capabilities: Image, Video, and Audio on the Frontier

{/* feishu-style:text-align:left */}
Accurate perception is the prerequisite for action. We benchmarked mimo-v2-omni against leading international models across all sensory modalities to ensure a rock-solid foundation for its capabilities as an AI agent.

{/* feishu-style:text-align:left */}
**Visual Understanding**: mimo-v2-omni demonstrates robust multidisciplinary visual reasoning and complex chart analysis. It has surpassed Claude 4.6 Opus and is rapidly closing the gap with top-tier closed-source models like Gemini 3.

{/* feishu-style:text-align:left */}
**Audio Understanding**: The model supports everything from environmental sound classification and multi-speaker separation to audio-visual joint reasoning and deep comprehension of continuous audio exceeding 10 hours. Its comprehensive performance exceeds Gemini 3 Pro, making it one of the most powerful audio understanding base models currently available.

{/* feishu-style:text-align:left */}
**Video Understanding**: By supporting native audio-video joint input, we have achieved true multimodal video comprehension. Through innovative video pre-training, the model possesses powerful situational awareness and predictive reasoning capabilities.

{/* feishu-style:text-align:left */}
When multiple modalities are processed simultaneously, the advantages of a unified architecture are magnified: cross-modal signals mutually reinforce one another rather than competing for resources.

## Agentic Capabilities: from Understanding to Execution

{/* feishu-style:text-align:left */}
If perception is the foundation, then action is the ultimate goal.

{/* feishu-style:text-align:left */}
A true AI agent model must be capable of observing complex environments across multiple modalities, formulating and executing plans, autonomously recovering from errors, and delivering end-to-end results.

### Omni-Modal Agent Tasks

{/* feishu-style:text-align:left */}
mimo-v2-omni excels in benchmarks involving interaction with real-world digital environments, performing on par with Gemini 3 Pro. This success is underpinned by its industry-leading perceptual capabilities：The more accurate the perception, the more effective execution.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/E1mxb9Phxohb4BxWxBXcnA3gnIf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/E1mxb9Phxohb4BxWxBXcnA3gnIf.png" width="642" height="233" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
At the same time, mimo-v2-omni remains highly competitive in text-only agent tasks.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DZu3bIDUWoBFyHxmBT5cZkGWnio.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/DZu3bIDUWoBFyHxmBT5cZkGWnio.png" width="636" height="295" alt="图片" style="margin: 16px auto;" />

## Capabilities Demonstration

### 💻 Browser-Use Scenarios

{/* feishu-style:text-align:left */}
Browser Use is the ultimate litmus test for a model’s agentic capabilities. It involves real-world interactions, dynamic web environments, heterogeneous interaction methods, and active anti-automation mechanisms. In these scenarios, the closed loop of perception, decision-making, and action operates continuously in an open environment until the mission is accomplished. When these same capabilities are ported to smart devices or robotics, they form the blueprint for General-Purpose Agents.

- **Shopping, Bargaining, and Ordering on Your Behalf**

   We tested an end-to-end shopping task. Controlling the browser, the model first browsed over a dozen posts on Xiaohongshu to complete information gathering and obtain purchasing recommendations. It then performed cross-platform price comparisons across multiple stores on JD, followed by connecting with human customer service to bargain using natural language. After real-time interaction with the representative, it ultimately completed the process of adding items to the cart and placing the order. The model autonomously handled non-standard DOM structures, multi-tab context management, and workflow recovery after encountering platform anti-automation detections.

- **TikTok Video Creation and Publishing**

   We tested an end-to-end video publishing task. The model autonomously designed four sets of visuals and synthesized all sound effects on-site with zero reliance on external assets. During rendering, it encountered a Chinese font error, which it self-corrected before continuing. It then controlled the browser to open the TikTok upload page, analyzed non-standard input controls to complete the copywriting, and proceeded to like and comment after clicking "Publish." Finally, it re-checked to confirm the video passed review and was publicly live.

### 🗒️ Smart Office Scenarios

{/* feishu-style:text-align:left */}
Through natural dialogue, mimo-v2-omni can directly generate high-quality Word documents, structured Excel sheets, professionally formatted PDFs, and complete PPTs. These generated documents are no longer drafts requiring heavy revision, but high-quality "near-final versions" tailored to actual needs.

- **2026 Intelligent College Entrance Examination Application**

   We tested the college entrance examination application planning task. The model can autonomously initiate web searches to obtain raw information, use skills to process files, and generate an Excel spreadsheet containing detailed application recommendations and tiered classifications.

## Open API

{/* feishu-style:text-align:left */}
The mimo-v2-omni model is now officially available via API with pricing:

- Input: $0.4 / million tokens;

- Output: $2 / million tokens.

{/* feishu-style:text-align:left */}
Visit [https://platform.xiaomimimo.com](https://platform.xiaomimimo.com/) to get started.


--- DOCUMENT: Xiaomi MiMo-V2-TTS: Versatile Voice Agent that Speaks and Sings ---
URL: https://mimo.mi.com/static/docs/news/previous-news/v2-tts-release.md

# Xiaomi MiMo-V2-TTS: Versatile Voice Agent that Speaks and Sings

{/* feishu-style:text-align:left */}
**Xiaomi mimo-v2-tts** is a large-scale speech synthesis model independently developed by Xiaomi. Built on a proprietary audio tokenizer and a multi-codebook joint speech–text modeling architecture, it has been trained on hundreds of millions of hours of speech data with large-scale pretraining and multi-dimensional reinforcement learning, enabling highly controllable, fine-grained speech style generation. mimo-v2-tts supports precise control ranging from global style setting to nuanced local emotional expression. It can perform tone shifts and gradual emotional transitions within a single utterance, faithfully reproducing the natural prosody of human speech. When singing, it can also accurately render pitch and rhythm, delivering natural and expressive performance.

{/* feishu-style:text-align:left */}
The mimo-v2-tts model is now available through the Xiaomi MiMo API open platform (https://platform.xiaomimimo.com), **with free access for a limited time**.

### Text Control
Flexible and customizable style control

{/* feishu-style:text-align:left */}
mimo-v2-tts supports free-form natural language descriptions instead of being limited to predefined keywords. The model can understand and follow arbitrary descriptive instructions.

- Emotion control: happy, sad, angry, gentle, excited, calm…

- Dialect support: Northeastern Mandarin, Sichuan dialect, Henan dialect, Cantonese, Taiwanese accent…

- Role play: Monkey King, Lin Daiyu, Iron Man…

- Freely combined phrases — true natural language control: “cute and coquettish, soft ‘baby voice’,” “lazy, just woke up, slightly husky,” “deeply affectionate, slow speaking pace,” “passionate and powerful”

### Fine-grained control of vocal events

{/* feishu-style:text-align:left */}
mimo-v2-tts can naturally insert and control various paralinguistic vocal events in speech, making the generated audio more realistic and expressive.

{/* feishu-style:text-align:left */}
Supported vocal events: laughter, coughing, pauses, thinking/hesitation, sighing, etc.

## Deep Text Understanding

{/* feishu-style:text-align:left */}
The model can intelligently recognize formatting cues in text and convert them into corresponding speech expressions—such as tone and punctuation—without requiring extra annotations.

{/* feishu-style:text-align:left */}
Format awareness → speech rendering:

- ALL CAPS text (e.g., “THIS IS IMPORTANT”) → automatically adds emphasis;

- Repeated words or characters (e.g., “no no no no no”) → automatically mapped to matching rhythm and emotion.

{/* feishu-style:text-align:left */}
During pretraining, the model learned from large-scale text–speech aligned data, enabling it to convert written formatting signals into natural-sounding speech.

## Beyond Speech: Dialects · Characters · Singing

{/* feishu-style:text-align:left */}
mimo-v2-tts goes beyond standard speech synthesis with rich and versatile expressive capabilities. It supports natural pronunciation across multiple dialects, enables role-playing with stylized character performances, and delivers high-quality singing synthesis—allowing a single model to speak, act, and sing with ease.

## Open API

{/* feishu-style:text-align:left */}
mimo-v2-tts is now officially available via API. **Free access is available for a limited time.** 

{/* feishu-style:text-align:left */}
Visit [https://platform.xiaomimimo.com](https://platform.xiaomimimo.com/) to get started.


--- DOCUMENT: MiMo-V2-Flash Release Note 2026/03/03 ---
URL: https://mimo.mi.com/static/docs/news/previous-news/news20260303.md

# MiMo-V2-Flash Release Note 2026/03/03

{/* feishu-style:text-align:left */}
mimo-v2-flash now supports web search, enabling access to real-time public information (such as news, products, weather, etc.).

{/* feishu-style:text-align:left */}
**Core Capabilities**

- **Flexible search modes**: Supports forced search and intent recognition. With intent recognition enabled, the model will autonomously decide whether to perform an online search without manual triggering.

- **Early search source return**: In the streaming response, the first packet will return all search sources.

- **Hybrid multi-tool invocation**: Can work with custom functions and tools; the model will automatically determine invocation priority and necessity.

- **Flexible response modes**: Supports both streaming and non-streaming responses, and both methods will return search and summary content.

{/* feishu-style:text-align:left */}
**Use Cases**

- **Real-Time News Aggregation**

   - Scenario: A user asks, "What are today's top stories about domestic large language models?"

   - Capability: The model automatically generates search keywords like "Chinese LLM latest news March 1 2026," searches the web, and returns a summarized response with source links.

- **Product Information & Price Comparison**

   - Scenario: A user asks, "What are the price and user reviews for the latest model of [Brand] phone?"

   - Capability: The model searches multiple e-commerce platforms for pricing and reviews, then organizes the information into a concise summary to aid decision-making.

- **Real-Time Weather & Travel Information**

   - Scenario: A user asks, "Is the weather in Shanghai tomorrow good for going out?"

   - Capability: The model fetches the Shanghai weather forecast and provides practical suggestions based on common sense, such as "Rain expected tomorrow in Shanghai, temperatures 10–15°C. Bring an umbrella and dress warmly."

{/* feishu-style:text-align:left */}
**Instructions and Recommendations**

1. **Enable Web Search Plugin**: Before using this feature, you need to activate the [Web Search Plugin](https://platform.xiaomimimo.com/#/console/plugin). For detailed parameters and invocation instructions, please refer to the [OpenAI API](https://platform.xiaomimimo.com/#/docs/api/text-generation/openai-api).

1. **Fees**: The web search feature incurs additional token consumption for generating search queries and processing results. A separate fee will also be charged per search call. For details, see [Web Search](https://platform.xiaomimimo.com/#/docs/usage-guide/tool-calling/web-search).


--- DOCUMENT: MiMo-V2-Flash Release Note 2026/02/04 ---
URL: https://mimo.mi.com/static/docs/news/previous-news/news20260212.md

# MiMo-V2-Flash Release Note 2026/02/04

1. **Upgraded Coding Capabilities in Thinking Mode:** 
Specifically optimized for programming scenarios, the Thinking Mode now achieves a score of **78.6** on SWE-Bench Verified. Both the resolution rate and the quality of code generation have been significantly improved.

1. **Substantial Boost in Tool Calling Accuracy:** 
Stability issues regarding tool usage have been resolved. Tool calling accuracy in Thinking Mode has surged from 64% to **97.0%** , greatly enhancing execution reliability in Agent scenarios.

1. **Enhanced Instruction Following & Reduced Hallucinations:** 

- **Instruction Following:**  Improved adherence to specific instructions, achieving an **AA-IFBench score of 72**.

- **Factuality:**  Enhanced rigor in factual responses, with the **Non-Hallucination Rate updated to 52%** .

1. **Optimized Handling of Complex Tasks:** 
Performance on Arena-Hard (Hard Prompts) in Thinking Mode has been strengthened, with the score rising to **60.6**. The model now demonstrates superior performance when handling high-difficulty logic problems.

1. **More Efficient Chain-of-Thought (CoT):** 
By optimizing CoT generation strategies, the consumption of redundant tokens has been significantly reduced. In benchmarks such as AIME25 and HMMT, the average generation length has decreased by **13% to 30%** . This effectively lowers latency and token costs while maintaining model performance.

<table>
<colgroup>
<col style="width: 34.41%" />
<col style="width: 23.82%" />
<col style="width: 23.82%" />
<col style="width: 17.950000000000003%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0204**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0112**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**SWE-Bench Verified** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**73.7**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">73.3</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">73.4</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**SWE-Bench Verified** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**78.6**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">74.2</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Hard Prompt)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**49.3**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">52.7</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">46.0</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Creative Writing)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**85.0**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">86.0</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">78.3</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Aren-Hard(Hard Prompt)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**60.6**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">58.3</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">54.1</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Creative Writing)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**85.8**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">90.4</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">86.2</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**AA-IFBench**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**72**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">64</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**AA-Omniscience Accuracy**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**19**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">27</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**AA-Omniscience Non-Hallucination Rate**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**52%** </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">9%</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Tool call success rate**</span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**97.0%** </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">64%</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">44%</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
  

<table>
<colgroup>
<col style="width: 24.92%" />
<col style="width: 15.02%" />
<col style="width: 15.02%" />
<col style="width: 15.02%" />
<col style="width: 15.02%" />
<col style="width: 15.000000000000014%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Benchmark**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash (Acc)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash (Avg Tokens)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0204 (Acc)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0204 (Avg Tokens)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Length Reduction Ratio (%)** </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**AIME25**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">94.8</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">26984</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">91.1</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">18879</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**30.04%** </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**HMMT_Feb_25**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">94.2</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">29294</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">92.9</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">21470</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**26.71%** </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**LiveCodeBench-AA**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">83.2</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">21488</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">84.9</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">18335</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**14.67%** </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**GPQA-Diamond**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">83.7</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">15862</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">83.8</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">13659</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**13.89%** </span></td>
</tr>
</tbody>
</table>

> Note: The model API call method and model name remain unchanged


--- DOCUMENT: Billing for Xiaomi MiMo API Open Platform is Launching Soon ---
URL: https://mimo.mi.com/static/docs/news/previous-news/billing.md

# Billing for Xiaomi MiMo API Open Platform is Launching Soon

{/* feishu-style:text-align:left */}
Dear Developers:

{/* feishu-style:text-align:left */}
Thank you for your attention and support for the Xiaomi MiMo API Open Platform. we hereby inform you that the platform will start billing at **Jan 26, 2026 16:00 (UTC+8)** , and before that, API calls will continue to be open for free, and the recharged amount or free credit will not be consumed for the time being. After the billing starts, **accounts with insufficient balance will be unable to use the API.** We recommend that you **recharge in advance** to avoid service interruption due to insufficient account balance.

### mimo-v2-flash API Pricing

- Domestic: Input ¥0.7 / 1M tokens, Input (Cache Hit) ¥0.07 / 1M tokens, Output ¥2.1 / 1M tokens

- Overseas: Input $0.1 / 1M tokens, Input (Cache Hit) $0.01 / 1M tokens, Output $0.3 / 1M tokens

### mimo-v2-flash API Pricing

- **China**: Input: ¥0.7 / 1M tokens, Cached Input: ¥0.07 / 1M tokens, Output: ¥2.1 / 1M tokens

- **Overseas**: Input: $0.1 / 1M tokens, Cached Input: $0.01 / 1M tokens, Output: $0.3 / 1M tokens

### Exclusive Benefits Available

{/* feishu-style:text-align:left */}
To thank you for your support, we have prepared exclusive free credits for all new and existing users. You can log in and go to [Balance](https://platform.xiaomimimo.com/#/console/balance) to collect them.

### Recharge Rules

1. **Real-Name Authentication**: Chinese mainland users must complete individual real-name authentication before recharging. Enterprise authentication is not yet available; please stay tuned for updates. Other users can recharge directly without real-name authentication.

1. **Recharge Process**: Go to the [Balance](https://platform.xiaomimimo.com/#/console/balance) page to recharge. Chinese mainland users can use Xiaomi Pay, Alipay, or WeChat Pay. Other users can use common payment methods such as Apple Pay, Google Pay, and credit/debit cards.

1. **Balance Inquiry**: Recharged funds are generally credited to your account in real time. You can view recharge orders and bonus amounts on the [Recharge Details](https://platform.xiaomimimo.com/#/console/recharge) page, and check your current balance on the [Balance](https://platform.xiaomimimo.com/#/console/balance) page. Once the billing system is activated, API calls will first deduct the complimentary free quota, followed by the recharged balance.

1. **Invoice Issuance**: Chinese mainland users can apply for electronic invoices by selecting successful recharge orders on the [Invoice](https://platform.xiaomimimo.com/#/console/invoice) page. Other users can download order receipts from the [Recharge Details](https://platform.xiaomimimo.com/#/console/recharge) page.

1. **For more details, please refer to** [FAQ](https://platform.xiaomimimo.com/#/docs/faq?target=payment-issue).

### Contact Us

{/* feishu-style:text-align:left */}
For assistance or business inquiries, feel free to reach out to us through the following channels:

- Email: support-mimo@xiaomi.com

- Scan the QR code at the bottom left to join the developer communication group

- Submit your feedback from [Contact Us](https://platform.xiaomimimo.com/#/contact)

{/* feishu-style:text-align:left */}
We are committed to continuously optimizing our products and services to provide you with superior AI capabilities. Thank you again for your understanding and trust!

{/* feishu-style:text-align:right */}
**—— Xiaomi MiMo API Open Platform Team**

{/* feishu-style:text-align:right */}
**January 23, 2026**


--- DOCUMENT: Xiaomi MiMo API Open Platform Recharge Function Officially Launched ---
URL: https://mimo.mi.com/static/docs/news/previous-news/recharge.md

# Xiaomi MiMo API Open Platform Recharge Function Officially Launched

## Dear Developers,

{/* feishu-style:text-align:left */}
Thank you for your continuous support of the Xiaomi MiMo API Open Platform. We are pleased to announce that the recharge service is now officially available as of **January 20, 2026**. To ensure the continuity of your service, we recommend completing your account recharge in advance to avoid any service interruptions due to insufficient balance after the billing system goes live.

### mimo-v2-flash API Pricing

- **China**: Input: ¥0.7 / 1M tokens, Cached Input: ¥0.07 / 1M tokens, Output: ¥2.1 / 1M tokens

- **Overseas**: Input: $0.1 / 1M tokens, Cached Input: $0.01 / 1M tokens, Output: $0.3 / 1M tokens

### Exclusive Benefits Available

{/* feishu-style:text-align:left */}
To thank you for your support, we have prepared exclusive free credits for all new and existing users. You can log in and go to [Balance](https://platform.xiaomimimo.com/#/console/balance) to collect them.

### About the Billing System

{/* feishu-style:text-align:left */}
Our billing system will officially launch in the near future. The exact date will be announced in an official announcement. Until then, API calls will remain free—your topped-up balance and free credits won’t be deducted for now.

### Recharge Rules

1. **Real-Name Authentication**: Chinese mainland users must complete individual real-name authentication before recharging. Enterprise authentication is not yet available; please stay tuned for updates. Other users can recharge directly without real-name authentication.

1. **Recharge Process**: Go to the [Balance](https://platform.xiaomimimo.com/#/console/balance) page to recharge. Chinese mainland users can use Xiaomi Pay, Alipay, or WeChat Pay. Other users can use common payment methods such as Apple Pay, Google Pay, and credit/debit cards.

1. **Balance Inquiry**: Recharged funds are generally credited to your account in real time. You can view recharge orders and bonus amounts on the [Recharge Details](https://platform.xiaomimimo.com/#/console/recharge) page, and check your current balance on the [Balance](https://platform.xiaomimimo.com/#/console/balance) page. Once the billing system is activated, API calls will first deduct the complimentary free quota, followed by the recharged balance.

1. **Invoice Issuance**: Chinese mainland users can apply for electronic invoices by selecting successful recharge orders on the [Invoice](https://platform.xiaomimimo.com/#/console/invoice) page. Other users can download order receipts from the [Recharge Details](https://platform.xiaomimimo.com/#/console/recharge) page.

1. **For more details, please refer to** [FAQ](https://platform.xiaomimimo.com/#/docs/faq?target=payment-issue).

### Contact Us

{/* feishu-style:text-align:left */}
For assistance or business inquiries, feel free to reach out to us through the following channels:

- Email: support-mimo@xiaomi.com

- Scan the QR code at the bottom left to join the developer communication group

- Submit your feedback from [Contact Us](https://platform.xiaomimimo.com/#/contact)

{/* feishu-style:text-align:left */}
We are committed to continuously optimizing our products and services to provide you with superior AI capabilities. Thank you again for your understanding and trust!

{/* feishu-style:text-align:right */}
**—— Xiaomi MiMo API Open Platform Team**

{/* feishu-style:text-align:right */}
**January 20, 2026**


--- DOCUMENT: MiMo-V2-Flash Release Note 2026/01/12 ---
URL: https://mimo.mi.com/static/docs/news/previous-news/news20260112.md

# MiMo-V2-Flash Release Note 2026/01/12

1. **Enhanced general capabilities:**  Improved the model’s performance on a wide range of general-purpose tasks.

1. **Upgraded coding performance in Thinking mode:**  Strengthened code generation quality in Thinking mode, especially for programming scenarios.

1. **Deep integration with Claude Code:**  Fully supports using Thinking mode in Claude Code.

   - **Best practice:**  Set Thinking as the default mode to achieve more stable, higher-quality code generation.

1. **Optimized Experience for Other Code Agents:**  Synchronized improvements to the interaction experience and generation quality across code assistant tools (Code Scaffolds) such as Kilo, Cline, and Roo.

1. **Improved Stability & Instruction Following:** Enhanced output stability and significantly improved adherence to specific output formats.

<table>
<colgroup>
<col style="width: 35.96%" />
<col style="width: 31.35%" />
<col style="width: 32.69%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0112**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**SWE-Bench Verified** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**73.3**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">73.4</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**SWE-Bench Verified Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**74.2**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Hard Prompt)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**52.7**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">46.0</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Creative Writing)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**86.0**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">78.3</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Hard Prompt)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**58.3**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">54.1</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Creative Writing)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**90.4**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">86.2</span></td>
</tr>
</tbody>
</table>

> Note: The model API call method and model name remain unchanged


--- DOCUMENT: MiMo Public Beta Free Access Extended ---
URL: https://mimo.mi.com/static/docs/news/previous-news/beta-free.md

# MiMo Public Beta Free Access Extended

{/* feishu-style:text-align:left */}
Dear Developers,

{/* feishu-style:text-align:left */}
Since the release of mimo-v2-flash, it has received widespread attention and enthusiastic support from users around the world. We appreciate your strong recognition and valuable feedback for the MiMo model! To reciprocate the enthusiastic support of our developers, we have decided to extend the public beta free trial period! The relevant adjustments are announced as follows:

### Extension of 20 days!

{/* feishu-style:text-align:left */}
The free trial period originally scheduled to end at the end of December 2025 will be postponed to **14:00 (Beijing Time) on January 20, 2026**. During this period, users can continue to use the mimo-v2-flash model API for free, utilizing its core features, including efficient text generation, code processing, etc. This extension is aimed at repaying the trust and support of our users, while encouraging more users to participate in testing and provide valuable feedback. We will further optimize model performance, enhance user experience, and ensure comprehensive security testing before the official release.

### Payment capability system launch preview

{/* feishu-style:text-align:left */}
The payment capability system of XiaomiMiMo API Platform will be enabled before the public beta ends and will undergo a trial run for about a week (the specific trial run period will be announced later). Users can top up their accounts during the trial period, but the service will not be charged at the time. Billing will start automatically and the balance will be consumed after the trial ends.To ensure normal use and stable service, we recommend that users recharge their accounts in advance.

{/* feishu-style:text-align:left */}
Attached is the model API pricing: **Domestic: Input: ¥0.7 / 1M tokens, Cached Input: ¥0.07 / 1M tokens, Output: ¥2.1 / 1M tokens;International: Input: $0.1 / 1M tokens, Cached Input: $0.01 / 1M tokens, Output: $0.3 / 1M tokens.** Please stay tuned to the official website announcements and Open Platform WeChat group notifications.

{/* feishu-style:text-align:left */}
We sincerely invite new and existing users to take advantage of the extended free public beta period to continue enjoying the intelligent experience brought by MiMo, and to actively share your usage experiences and suggestions. If you have any questions, please feel free to [contact us](https://platform.xiaomimimo.com/#/contact) or join the developer discussion group (scan the QR code in the lower left corner to join). Let's explore the boundaries of intelligence together.

{/* feishu-style:text-align:right */}
**—— Xiaomi MiMo API Open Platform Team**


--- DOCUMENT: MiMo-V2-Flash Release 2025/12/16 ---
URL: https://mimo.mi.com/static/docs/news/previous-news/news20251216.md

# MiMo-V2-Flash: High-Efficiency Inference, Code & Agent Foundation Model

{/* feishu-style:text-align:left */}
Xiaomi mimo-v2-flash is officially open-sourced today!!! It is a MoE (Mixture of Experts) model designed for extreme inference efficiency, with 309B total and 15B activated parameters. 

{/* feishu-style:text-align:left */}
With an innovative hybrid attention and multi-layer MTP (Multi-Token Prediction) architecture, the model ranks **top 2 among all open-source models** across multiple agent evaluation benchmarks. Besides, its coding capability surpasses all open-source models and is comparable with Claude 4.5 Sonnet, while its **inference cost is only 2.5%**  of Claude’s and its **generation speed is doubled** — pushing inference efficiency to the limit.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YySMbAqyyoOKG6xVdbScif2UnYp.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YySMbAqyyoOKG6xVdbScif2UnYp.png" width="656" height="313" alt="图片" style="margin: 16px auto;" />

{/* feishu-style:text-align:left */}
To foster open-source engagement, both the model weights and inference code are fully open-sourced under the MIT license. 

{/* feishu-style:text-align:left */}
**The API is available free of charge for a limited time.** 

## Extreme Optimization of Cost and Speed

{/* feishu-style:text-align:left */}
The API pricing for mimo-v2-flash is **$0.1 per million input tokens and $0.3 per million output tokens.** 

{/* feishu-style:text-align:left */}
In the chart below, the horizontal axis compares speed and cost across leading models——mimo-v2-flash achieves both the lowest cost and the highest speed.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/N2uQbZmpaoHjhHxPZtjcjAz0nTf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/N2uQbZmpaoHjhHxPZtjcjAz0nTf.png" width="616" height="380" alt="图片" style="margin: 16px auto;" />

## Architectural innovations designed for high-efficiency inference

{/* feishu-style:text-align:left */}
Key architectural designs: 

- **Hybrid Attention:** We adopt a hybrid attention mechanism combining Global Attention and Sliding Window Attention (SWA) at a 1:5 ratio, where the window size of SWA is 128. During pre-training, we train the model with a context length of 32k, and extend it to 256k. Compared to mainstream Linear Attention approaches, extensive early-stage empirical studies show that SWA is simple, efficient, and practical, delivering stronger overall performance in general tasks, long-context handling, and reasoning. It also provides a fixed-size KV cache, making it easy to integrate with existing training and inference infrastructure.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YcY9b26lOo8vdFxRcXMcStthnRe.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/YcY9b26lOo8vdFxRcXMcStthnRe.png" width="528" height="453" alt="图片" style="margin: 16px auto;" />

- **MTP Inference Acceleration**: We introduce MTP (Multi-Token Prediction) to strengthen the model's capability and speed up inference. During inference, MTP validates MTP tokens in parallel, breaking the memory bandwidth bottleneck of traditional decoding under large batch sizes. In practice, a 2.5×–3.7× real-world speedup is achieved with a 3-layer MTP setup.

<img src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/LGrtb2E7Soz2stxEoNpc9ucAnFf.png" data-preview-src="https://aistudio-cdn.xiaomimimo.com/xiaomimimo-static/mimo-docs-figures/display/LGrtb2E7Soz2stxEoNpc9ucAnFf.png" width="566" height="377" alt="图片" style="margin: 16px auto;" />

## Related links

- Technical Report：[https://github.com/XiaomiMiMo/MiMo-V2-Flash/blob/main/paper.pdf](https://github.com/XiaomiMiMo/MiMo-V2-Flash/blob/main/paper.pdf)

- Model Weights：https://hf.co/XiaomiMiMo/MiMo-V2-Flash

- Github Repository：https://github.com/xiaomimimo/MiMo-V2-Flash

- Blog Post: https://mimo.xiaomi.com/blog/mimo-v2-flash

- LMSYS Blog：[https://lmsys.org/blog/2025-12-16-mimo-v2-flash](https://lmsys.org/blog/2025-12-16-mimo-v2-flash/)


--- DOCUMENT: Model Release ---
URL: https://mimo.mi.com/static/docs/updates/model.md

# Model Release

## 2026-09-22 MiMo-V2.6 Series Release

- **mimo-v2.6-pro：** Our most powerful flagship reasoning model — omni-modal, ultra-high performance, trillion-parameter — built for complex projects, long-horizon tasks, high-stakes work, cybersecurity, and research.

- **mimo-v2.6-flash：** Full-modality, high-intelligence, low-cost reasoning model — the best balance for high-frequency calls and large-scale tasks in professional workflows.

- **mimo-v2.6-pro-ultraspeed：** Flagship V2.6-Pro performance, up to 20x faster. Built for real-time and latency-sensitive workloads.

## 2026-06-02 mimo-v2.5-asr Released

{/* feishu-style:text-align:left */}
Model Introduction:

- **Bilingual & Dialects:** Supports Chinese, English, code-switching, and various regional dialects (Wu, Cantonese, Minnan, Sichuanese).

- **Lyrics Transcription:** High-accuracy Chinese/English lyrics transcription in mixed vocal-instrumental tracks.

- **Robust in Complex Audio:** Excels in challenging environments (high noise, far-field, multi-speaker).

- **Knowledge-Intensive AI:**  Pinpoint accuracy for classical poetry, jargon, and proper nouns, with auto-punctuation.

## 2026-04-23 mimo-v2.5-pro Released

{/* feishu-style:text-align:left */}
Model Introduction:

- **Trillion parameters, efficient architecture:** 1T total parameters | 42B activations | 1M ultra-long context

- **Ultimate Agent Performance:** In high-intensity agent scenarios, it performs comparably to Claude Opus4.6

## 2026-04-23 mimo-v2.5 Released

{/* feishu-style:text-align:left */}
Model Introduction:

- **Native full-modal perception + 1M context:**  Supports native understanding of images, videos, audio, and text, enabling cross-modal precise perception and long-range reasoning, with comprehensive perception capabilities ranking among the industry's forefront 

- **Powerful full-modal Agent capabilities:**  It has native Agent execution capabilities, enabling it to efficiently complete complex tasks such as browsing, understanding, reasoning, and operation, with its performance in daily tasks comparable to that of **mimo-v2.5-pro**

- **Combining Performance and Efficiency:** While maintaining leading capabilities, achieving superior token efficiency, and positioned at the Pareto frontier of performance and efficiency

## 2026-04-23 MiMo-V2.5-TTS Series Release

{/* feishu-style:text-align:left */}
Model Introduction:

- **Premium Voice TTS:**  Built-in with multiple high-quality premium voices, it has strong capabilities in understanding and adhering to style instructions, supports fine-grained control over speech rate, emotion, tone, etc., and meets the expression needs of multiple scenarios 

-  **Timbre Design:** Supports quickly defining and generating new timbres through a single sentence, making timbre creation more intuitive and efficient

-  **Timbre Cloning:** Based on a small number of audio samples, it can reproduce the target timbre with high fidelity, while maintaining the consistency of timbre characteristics and possessing good generalization and stability

## 2026-03-18 mimo-v2-pro Release

{/* feishu-style:text-align:left */}
**Model overview:** 

- Uses hybrid architecture with a 1:7 ratio of Global Attention to Sliding Window Attention (SWA);

- 1T total parameters, with 42B active parameters;

- Supports an ultra-long context window of 1M tokens.

{/* feishu-style:text-align:left */}
**Model details:**  https://platform.xiaomimimo.com/#/docs/news/v2-pro-release

## 2026-03-18 mimo-v2-omni Release

{/* feishu-style:text-align:left */}
**Model overview:** 

- Supports up to 256K context length;

- Supports text, vision, and speech modalities.

{/* feishu-style:text-align:left */}
**Model details:**  https://platform.xiaomimimo.com/#/docs/news/v2-omni-release

## 2026-03-18 mimo-v2-tts Release

{/* feishu-style:text-align:left */}
**Model overview:** 

- Pretrained on over 100 million hours of data, using a self-developed multi-codebook speech modeling architecture;

- Offers unique capabilities such as style control, singing, and voice cloning.

{/* feishu-style:text-align:left */}
**Pricing:**  free for a limited time.

{/* feishu-style:text-align:left */}
**Model details:** https://platform.xiaomimimo.com/#/docs/news/v2-tts-release

## 2026-02-04 mimo-v2-flash Update

1. **Upgraded Coding Capabilities in Thinking Mode:** 
Specifically optimized for programming scenarios, the Thinking Mode now achieves a score of **78.6** on SWE-Bench Verified. Both the resolution rate and the quality of code generation have been significantly improved.

1. **Substantial Boost in Tool Calling Accuracy:** 
Stability issues regarding tool usage have been resolved. Tool calling accuracy in Thinking Mode has surged from 64% to **97.0%** , greatly enhancing execution reliability in Agent scenarios.

1. **Enhanced Instruction Following & Reduced Hallucinations:** 

- **Instruction Following:**  Improved adherence to specific instructions, achieving an **AA-IFBench score of 72**.

- **Factuality:**  Enhanced rigor in factual responses, with the **Non-Hallucination Rate updated to 52%** .

1. **Optimized Handling of Complex Tasks:** 
Performance on Arena-Hard (Hard Prompts) in Thinking Mode has been strengthened, with the score rising to **60.6**. The model now demonstrates superior performance when handling high-difficulty logic problems.

1. **More Efficient Chain-of-Thought (CoT):** 
By optimizing CoT generation strategies, the consumption of redundant tokens has been significantly reduced. In benchmarks such as AIME25 and HMMT, the average generation length has decreased by **13% to 30%** . This effectively lowers latency and token costs while maintaining model performance.

<table>
<colgroup>
<col style="width: 34.41%" />
<col style="width: 23.82%" />
<col style="width: 23.82%" />
<col style="width: 17.950000000000003%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0204**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0112**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**SWE-Bench Verified** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**73.7**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">73.3</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">73.4</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**SWE-Bench Verified** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**78.6**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">74.2</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Hard Prompt)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**49.3**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">52.7</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">46.0</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Creative Writing)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**85.0**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">86.0</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">78.3</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Aren-Hard(Hard Prompt)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**60.6**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">58.3</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">54.1</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Creative Writing)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**85.8**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">90.4</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">86.2</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**AA-IFBench**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**72**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">64</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**AA-Omniscience Accuracy**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**19**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">27</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**AA-Omniscience Non-Hallucination Rate**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**52%** </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">9%</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Tool call success rate**</span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**97.0%** </span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">64%</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">44%</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
  

<table>
<colgroup>
<col style="width: 24.92%" />
<col style="width: 15.02%" />
<col style="width: 15.02%" />
<col style="width: 15.02%" />
<col style="width: 15.02%" />
<col style="width: 15.000000000000014%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Benchmark**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash (Acc)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash (Avg Tokens)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0204 (Acc)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0204 (Avg Tokens)** </span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**Length Reduction Ratio (%)** </span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**AIME25**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">94.8</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">26984</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">91.1</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">18879</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**30.04%** </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**HMMT_Feb_25**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">94.2</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">29294</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">92.9</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">21470</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**26.71%** </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**LiveCodeBench-AA**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">83.2</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">21488</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">84.9</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">18335</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**14.67%** </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**GPQA-Diamond**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">83.7</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">15862</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">83.8</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">13659</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**13.89%** </span></td>
</tr>
</tbody>
</table>

> Note: The model API call method and model name remain unchanged

## 2026-01-12 mimo-v2-flash Update

1. **Enhanced general capabilities:**  Improved the model’s performance on a wide range of general-purpose tasks.

1. **Upgraded coding performance in Thinking mode:**  Strengthened code generation quality in Thinking mode, especially for programming scenarios.

1. **Deep integration with Claude Code:**  Fully supports using Thinking mode in Claude Code.
- **Best practice:**  Set Thinking as the default mode to achieve more stable, higher-quality code generation.

1. **Optimized Experience for Other Code Agents:**  Synchronized improvements to the interaction experience and generation quality across code assistant tools (Code Scaffolds) such as Kilo, Cline, and Roo.

1. **Improved Stability & Instruction Following:** Enhanced output stability and significantly improved adherence to specific output formats.

<table>
<colgroup>
<col style="width: 35.96%" />
<col style="width: 31.35%" />
<col style="width: 32.69%" />
</colgroup>
<thead>
<tr>
<th></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash-0112**</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">**mimo-v2-flash**</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**SWE-Bench Verified** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**73.3**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">73.4</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**SWE-Bench Verified Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**74.2**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">-</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Hard Prompt)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**52.7**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">46.0</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Creative Writing)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Non-Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**86.0**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">78.3</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Hard Prompt)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**58.3**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">54.1</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">**Arena-Hard(Creative Writing)** </span><br /><span style="display: inline-block; width: 100%; text-align: left;">**Thinking**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**90.4**</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">86.2</span></td>
</tr>
</tbody>
</table>

> Note: The model API call method and model name remain unchanged

## 2025-12-16 mimo-v2-flash Release

{/* feishu-style:text-align:left */}
**Model overview:** 

- Uses hybrid architecture with a 1:5 ratio of Global Attention to Sliding Window Attention (SWA), a window size of 128, native 32K context, and extended training up to 256K;

- Introduces 3 MTP layers, delivering 2.5 to 3.7× faster inference.

{/* feishu-style:text-align:left */}
**Pricing:**  input 0.1/M tokens, output 0.3/M tokens.

{/* feishu-style:text-align:left */}
**Model details:**  [mimo-v2-flash: High-Efficiency Inference, Code & Agent Foundation Model](https://platform.xiaomimimo.com/#/docs/news/news20251216)

{/* feishu-style:text-align:left */}
**Usage guide:**  [First API call](https://platform.xiaomimimo.com/#/docs/quick-start/first-api-call)


--- DOCUMENT: Model Deprecation ---
URL: https://mimo.mi.com/static/docs/updates/deprecate.md

# Model Deprecation

{/* feishu-style:text-align:left */}
With the continuous iteration of the MiMo model, the new version has comprehensively outperformed the old version in terms of effectiveness and performance. We will gradually deprecate the legacy models, and the specific plan will be announced in advance via SMS, email, website announcements, etc. Please pay attention to the relevant messages and complete the switch in a timely manner.

{/* feishu-style:text-align:left */}
**Time Definition**

- System replacement time: The time when the offline model is automatically switched to the new version model. After this time, requests using the old version model name will be automatically replaced with the corresponding new version model and billed according to the new version model. [View Pricing](https://platform.xiaomimimo.com/docs/zh-CN/price/pay-as-you-go)

- Deprecation Time: The time when the name of the old model expires. After this time, requests using the name of the old model version will receive an error message. Please ensure that the model replacement is completed before this time.

{/* feishu-style:text-align:left */}
**Operational Recommendations** 

- Access [ Bill Details ](https://platform.xiaomimimo.com/console/usage), check if there are any models pending offline; 

- Refer to the system replacement model in the table below to complete your code self-check and replacement. It is recommended to fully test and verify before the official switch. 

{/* feishu-style:text-align:left */}
<br></br>

### Deprecated model on 2026.10.21

<table>
<colgroup>
<col style="width: 15.08%" />
<col style="width: 22.05%" />
<col style="width: 62.87%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Offline Model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Deprecated Time</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Note</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.10.21 10:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**No system replacement model; will be directly deprecated upon expiration**</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.10.21 10:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">**No system replacement model; will be directly deprecated upon expiration**</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
<br></br>

### Deprecated model on 2026.6.30 

<table>
<colgroup>
<col style="width: 14.88%" />
<col style="width: 21.76%" />
<col style="width: 21.76%" />
<col style="width: 13.56%" />
<col style="width: 28.039999999999992%" />
</colgroup>
<thead>
<tr>
<th><span style="display: inline-block; width: 100%; text-align: left;">Deprecated Model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Deprecated Time</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">System replacement time</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">System Replacement Model</span></th>
<th><span style="display: inline-block; width: 100%; text-align: left;">Replacement Impact</span></th>
</tr>
</thead>
<tbody>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.6.30 00:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.6.1 00:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-pro</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">API parameters are fully adapted</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2-omni</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.6.30 00:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.6.1 00:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">API parameters are fully adapted</span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2-flash</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.6.30 00:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.6.18 00:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">The default value of the parameter has changed, see details below </span></td>
</tr>
<tr>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2-tts</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.6.30 00:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Beijing Time 2026.6.27 00:00</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">mimo-v2.5-tts</span></td>
<td><span style="display: inline-block; width: 100%; text-align: left;">Timbre remapping,`mimo_default` is mapped to `冰糖` in Chinese clusters and `mia` in other clusters.</span></td>
</tr>
</tbody>
</table>

{/* feishu-style:text-align:left */}
**Note:** 

{/* feishu-style:text-align:left */}
Starting from 00:00 on June 18, 2026, Beijing time, requests for mimo-v2-flash will be automatically routed to mimo-v2.5. The relevant parameter processing rules are as follows: 

- mimo-v2.5 does not support customizing `temperature` and `top_p` in thinking mode. The actual parameters passed to the model are `temperature: 1.0` and `top_p: 0.95`

- If parameters are customized when using mimo-v2-flash, the parameters passed to mimo-v2-flash will be inherited when automatically routing to mimo-v2.5 

- If ` thinking `, ` temperature `, or ` max_completion_tokens ` are not specified in the request, the system will automatically  use  ` mimo-v2.5 ` default values , as shown in the following table:


--- DOCUMENT: Xiaomi MiMo API Platform ---
URL: https://mimo.mi.com/static/docs/updates/feature/platform.md

# Xiaomi MiMo API Platform 

## 2026-9-22 Batch Inference Now Available

{/* feishu-style:text-align:left */}
Batch Inference is an offline solution for processing large volumes of data that don't require real-time responses. With an OpenAI-compatible API, it's ideal for use cases such as model evaluation, data labeling, and batch regression testing. You can create asynchronous tasks via the SDK or the console.

## 2026-9-15 In-App Messages Now Available

{/* feishu-style:text-align:left */}
Platform announcements and system notifications are now delivered  in real time via in-app messages, ensuring that no critical updates go unnoticed.

## 2026-6-23 Support for OpenAI Responses API

{/* feishu-style:text-align:left */}
Supports OpenAI Responses API-compatible endpoints, enabling integration with coding tools such as Codex.

## 2026-6-11 Refer & earn New Upgrade

{/* feishu-style:text-align:left */}
Invite friends to buy their first Token Plan. You get cash back, and they get a first-order discount.

## 2026-4-29 Refer & earn New Release

{/* feishu-style:text-align:left */}
Invite builders to try Xiaomi's flagship MiMo V2.5 — sign up to instantly get $2 in trial credits, redeemable for API calls (Token Plan not supported).

## 2026-4-23 Token Plan Auto-Renewal Now Available

- Monthly and annual subscription plans available, with options to renew your current plan or upgrade to a new one.

- Five new models added to all plans: **mimo-v2.5-pro、mimo-v2.5、mimo-v2.5-tts-voiceclone、mimo-v2.5-tts-voicedesign、mimo-v2.5-tts.** 

- Tiered token consumption based on context length has been removed for all models (previously 2x consumption for 256K–1M context).

- Multiple limited-time promotions are now available:

- **Returning user bonus** : exclusive "usage reset" reward for existing users.

- **First-time auto-renewal discount** : new users who have never subscribed get 23% off; returning users enabling auto-renewal for the first time get 30% off. This discount cannot be combined with the first-purchase discount. Limited to one per account.

- **Annual plan savings** : 12% off compared to monthly auto-renewal. First-purchase and first-time auto-renewal discounts do not apply to annual plans.

- **Off-peak discount rate** : 0.8x token consumption during off-peak hours (12:00 AM – 8:00 AM Beijing Time / 4:00 PM – 12:00 AM UTC).

## 2026-4-3 Token Plan New Release

- Offers four tiers of packages, a unified Credit quota system, calculates consumption based on Token usage, and supports package upgrades and usage queries 

- Flexibly call mimo-v2-pro / omni / tts, sharing quota

- Compatible with mainstream AI programming tools: OpenClaw | Claude Code | OpenCode | Kilocode | Cline... 

## 2026-03-03 mimo-v2-flash Update

{/* feishu-style:text-align:left */}
**Model Plugin Capabilities**

- mimo-v2-flash supports Web Search tool to access real-time public network information

## 2026-02-10 Real-name authentication of enterprises and corporate payment functions are launched 

{/* feishu-style:text-align:left */}
**Real-name authentication of enterprise accounts and  domestic non-mainland users**

- Added real-name authentication for enterprises

- Added real-name authentication for domestic non-mainland users

- Changed the account that supports real-name authentication to an enterprise account

{/* feishu-style:text-align:left */}
**B2B corporate online banking top-up**

- Corporate accounts support payment through corporate online banking

- Support corporate order refunds and invoicing

## 2026-01-26 Billing Function Launched

{/* feishu-style:text-align:left */}
**Model inference service billing**

- The model inference service is billed based on token consumption, and supports viewing and downloading bills and checking real-time balance.

## 2026-01-20 The Recharge Function Launched

{/* feishu-style:text-align:left */}
**Recharge and invoicing**

- Domestic users: Personal real-name authentication.

- Global users: Recharge, refund, invoicing, balance alerts, metering and billing, and billing.

{/* feishu-style:text-align:left */}
**Bill details**

- Existing users' free credit will be automatically credited, and new users will receive free credit upon registration.

## 2025-12-16 Xiaomi MiMo API Open Platform Released

{/* feishu-style:text-align:left */}
**API Key Management and Usage Tracking**

- Supports quick creation and configuration of API keys in the console to achieve model call authentication.

- Real-time viewing and export of usage data such as token consumption and call count.

{/* feishu-style:text-align:left */}
**One-stop developer documentation center**

- Provides a quick guide to getting started with the platform, model API documentation, and integration extension methods

- Includes model introduction, debugging examples, and troubleshooting scenarios


--- DOCUMENT: Xiaomi MiMo Desktop ---
URL: https://mimo.mi.com/static/docs/updates/feature/desktop.md

# Xiaomi MiMo Desktop

## V0.1.0（2026-09-01）

{/* feishu-style:text-align:left */}
The first version of Xiaomi MiMo Desktop delivers enhanced task orchestration, intelligent scheduling and full-modal working capabilities.

#### New Features

- **Collaborative Optimization of Model and Harness** ：Deeply optimized for the MiMo series of models to enhance their capabilities in tool invocation, context management, and complex task execution.

- **Intelligent Scheduling and Collaboration**: Smart Router automatically identifies the type and complexity of a task, and matches it with the appropriate models, Harness and dedicated agents.

- **Cross-session collaboration**: Multiple sessions can share objectives, context and task progress, supporting multi-role consultation, task handover and automated collaboration workflows.

- **Full-modal Input and Output**: Supports the understanding, generation and editing of documents, images, audio, videos and Compressed Packet.

- **In-conversation Content Display**: Videos and images can be previewed directly; data, charts and documents support sidebar display, and interactive components can also be generated quickly.

- **end-to-end editing collaboration**: Office files and HTML support high-fidelity preview, partial editing and version rollback, and can be further modified after generation until delivery.

- **Stable browser and computer control performance**: autonomously read and operate webpages to complete information retrieval, form filling, creatives extraction and webpage verification; independently operate desktop applications (overseas version only) to achieve cross-application collaboration, and reuse operation processes via Record & Replay.

#### Performance and Efficiency

- **Intelligent Scheduling**: Automatically balances task effectiveness, response speed and usage cost.

- **High cache reuse**: the cache hit rate can reach up to **99%** for the same session, and up to **95%** across sessions.

- **Local generation**: Only regenerate the content that needs to be modified, which saves tokens and shortens the waiting time.

{/* feishu-style:text-align:left */}
MiMo Desktop empowers complex tasks to evolve from "merely getting done" to being well-planned, collaboratively executed, and continuously optimized.

{/* feishu-style:text-align:left */}
Download：https://mimo.xiaomimimo.com/desktop/


--- DOCUMENT: Xiaomi MiMo Claw ---
URL: https://mimo.mi.com/static/docs/updates/feature/claw.md

# Xiaomi MiMo Claw

## V2.2.0（2026-09-21）

{/* feishu-style:text-align:left */}
**Supports the MiMo V2.6 model**

- Update the main inference model to mimo-v2.6-pro

- Updated the visual understanding skill to mimo-v2.6-flash

## V2.1.5（2026-09-10）

- Added storage usage monitoring and usage cleanup reminders

- Optimize the data export speed and improve the success rate of data derivation

- Fixed the issue where quota does not take effect in specific scenarios

## V2.1.4（2026-09-03）

- Adjust the duration of a single free trial to 3 hours

- Optimize the security risk control identification logic to reduce the misoperation of pods

- Fix issues such as the failure of storage upper limit control

## V2.1.3（2026-08-27）

- Optimize some context management strategies to improve the success rate of dialogue execution

- Optimize the real-time logic of the network tool

- Fix the availability issue of showcase file storage

## V2.1.2（2026-08-18）

- Optimize the security protection of file links, enhance the link security of uploaded and conversation-generated files, and further safeguard file privacy and security

- Adjust the single-session experience duration of the free version

- Fix the cluster configuration error issue

## V2.1.1（2026-08-13）

- Add a cap on the computing power of the new one-time free version, and implement usage monitoring and front-page prompts

## V2.1.0（2026-07-28）

- Add a series of risk control strategies to continuously reduce the probability of misjudging normal users

## V2.0.6（2026-07-21）

- Optimize file download time consumption to enhance the file export experience of Workspace

## V2.0.5（2026-06-26）

- Optimize the content of SOUL. md, and flexibly adjust the scope of security control to facilitate users in performing more tasks

- Optimize the adaptive display of the file export button in browsers such as Firefox

## V2.0.4（2026-06-25）

- Supports title modification, automatic generation, and messaging Channel titles are fixed titles to improve task Dialog Management efficiency

- Adjust the free data retention period from 2 hours to 24 hours

- Fixed the issue where the display time and data retention time of the export entry (subscription + free) on the home page are inconsistent

## V2.0.3（2026-06-24）

- After optimizing the saving of Kingsoft Office documents, you can view the latest content in real time

- Optimize the speed of generating QR codes in WeChat

## V2.0.2（2026-06-23）

- Optimize the operation speed of Kingsoft Docs during preview/editing

- Fixed the timeout error when establishing connection via QQ and Feishu channels

## V2.0.1（2026-06-22）

- Optimized the stability of task loading, pop-up window link supplementation and other aspects, and solved the problem that manual modification of OpenClaw configuration and environment variables was required

- Update the openclaw version to fix the cache issue caused by the logical sequence of context, thereby optimizing the problem of excessive credit consumption

- Added validation for the filling format of the new API key and interactive prompts

- New Product Positioning Description for Home Page

## V2.0.0（2026-06-16）

{/* feishu-style:text-align:left */}
**MiMo Claw: Official Monetization Version Released**

- Added adaptation to the four tiers of annual/monthly subscriptions of tokenplan, supporting one-click subscription check; and also supports filling in the api key to associate with Platform for redirection and usage

- Deepen cooperation with Kingsoft Office to realize a one-stop closed loop covering document generation, preview, editing and export

- Added more convenient operations, such as messages channels, Skills, etc.

- Added support for multi-conversation and multi-task parallel processing

- Newly added file management supports batch download

- Added exception handling capabilities, such as restart, automatic repair, and factory reset

- New Subscription Benefit Information: Overview of Legal Terms and Subscription Benefits, etc.

- Optimize page styles and enhance interaction experience

## V1.1.0（2026-04-17）

- Equipped with the latest 2.5 series models: MiMo-V2.5-Pro, MiMo-V2.5, and MiMo-V2.5-TTS

## V1.0.3（2026-04-04）

- Add content moderation policies for document preview

## V1.0.2（2026-03-25）

- Fixed issues such as the malfunction of the Feishu plugin in the Claw chat room

- Fix exceptions such as file preview

## V1.0.1（2026-03-20）

- Optimize the security verification of mimoclaw key

## V1.0.0（2026-03-19）

{/* feishu-style:text-align:left */}
**MiMo Claw: Beta Free Version Launched Globally**

- Provide a free cloud-based shrimp product experience with a limited single session duration

- User isolation throughout the entire lifecycle of cloud services: from creation, isolated usage to decommissioning

- Convenient workspace display, supporting quick viewing of workspace files and tool scheduling progress

- Document generation supports Kingsoft Office online preview

- High-quality skill presets, curated showcase usage guides

- MiMo Claw Chat Room


--- DOCUMENT: Xiaomi MiMo Studio ---
URL: https://mimo.mi.com/static/docs/updates/feature/studio.md

# Xiaomi MiMo Studio

## V3.4.0（2026-09-21）

- **Added entry for UltraSpeed high-speed inference model: flagship capabilities, ultra-fast response.** 

- **New MiMo V2.6 model experience available**

   - It supports the MiMo Chat module to experience the mimo-v2.6-flash and mimo-v2.6-pro models for dialogue, and is compatible with multimodal visual understanding functions.

   - showcase and display interaction upgrade

   - Optimize system prompts

- Added a new version update entry to record product iteration details

- Fix some security vulnerabilities

## V3.3.3（2026-08-27）

- Optimize the playback visualization animation effects for audio generation

- Optimize the shareable logic of sessions

- Fixed the display icon for dialogue consumption

## V3.3.2（2026-08-13）

- Optimize the security protection of file links across all platforms, enhance the link security of uploaded and conversation-generated files, and further safeguard file privacy

- Optimize the recognition accuracy of mixed-modal uploads

## V3.3.1（2026-08-11）

- Added login status synchronization for all MiMo web-based products, covering MiMo Studio / MiMo Claw / MiMo Platform / MiMo UltraSpeed / official homepage

- Optimize the identification of online search intentions and optimize brand identification strategies

- Adjust the call logic of content moderation

- Optimize the scrolling interactive display of the conversation directory in the two areas of claw and chat

## V3.3.0（2026-07-23）

- Full-platform UI redesign: Two visual styles, "Rational Humanistic" and "Dark Tech", are provided to create an AI service experience that is closer to users

   - Open Platform Redirection: Unifying Brand Recognition and Visual Consistency

   - Speech Model Module: Making AI Avatars More Approachable

   - Multimodal Interaction Page: Improving the Quality of Interaction Details

   - Integrated settings page for a cleaner interface

- Optimize the display strategy of new arrival pop-up banners to reduce interference to users

- Fix the display logic of the multimodal Showcase

## V3.2.0（2026-07-05）

- New MiMo ASR model experience module added

- Optimize the showcase interaction experience of 4 voice modules

- Optimize the audio speech selection and recognition strategy

- Added audio download security generation prompt splicing, and optimized the blank time to make the splicing smoother

## V3.1.3（2026-06-07）

- Optimize file upload and extend support for doc/docx formats

- Temporarily disable the photo-based problem-solving function (during the National College Entrance Examination and Senior High School Entrance Examination period)

- Retire the MiMo-V2.0 series models and their dependent capabilities

- Update domestic network-connected data vendors

## V3.1.2（2026-05-14）

- Optimize the logic for sorting and displaying conversation history

- Optimize the platform rate limiting strategy

- Optimize file parsing and validation logic

- Fixed the issue where an error occurs when sharing to continue a conversation

- Fixed the issue with sharing session permissions

## V3.1.1（2026-05-11）

- New voice download function, supporting one-click export of audio generated by 3 voice large models

- Adjust the timbre descriptions and broadcast text caps of 3 voice large language models, and expand both to 1024 characters

- Optimize platform entry records to remember the last selection used

- Optimize the Cantonese broadcast effect of TTS

- Optimize security hardening and operation documents

## V3.1.0（2026-04-24）

- Full-platform update for the MiMo V2.5 series models

- Supports experience of 3 voice models

   - Speech Generation Model: 8 Premium Voices, Multiple Broadcast Modes

   - Timbre Design Model: Supports long-text timbre description, and newly adds AI-optimized style description and broadcast text

   - Timbre Cloning Model: Supports online recording and local file upload, enabling one-click timbre cloning

   - Supports Chinese-English bilingualism, and is compatible with PC and mobile devices

- Added a platform operation banner for the promotion of new releases on the platform

- Update the legal privacy agreement and supplement the voice-generated content

## V3.0.1（2026-04-17）

- Fix the display logic of the user agreement popup

- Fixed the issue caused by the adjustment of the internal gateway domain name

## V3.0.0（2026-04-08）

{/* feishu-style:text-align:left */}
Added the MiMo-V2-TTS model experience module to facilitate quick testing of the model's text-to-speech broadcasting capabilities

- Four-mode voice broadcast

- Display Adaptation: Dark/Light Theme Colors, Chinese/English, PC/Mobile Devices

- Operations Feedback: TTS User Communication Portal & Exclusive User Group

- Content Security: Enhanced review with newly added input

## V2.0.0（2026-03-19）

- Added the experience module for the first new flagship model MiMo-V2-Pro

- Newly added experience module for the visual understanding model MiMo-V2-Omni

   - Support multi-modal input

   - It supports application scenarios such as multilingual translation, photo-based problem solving, and smart photo question answering.

   - Support multi-modal file upload

## V1.3.3（2026-03-02）

- Fixed issues such as the differentiation of account ban status

## V1.3.2（2026-02-26）

- Update the filing number of Xiaomi MiMo model service, the webpage announcement of the Cyberspace Administration of China and other relevant information

## V1.3.1（2026-02-05）

- Retain user input content when switching login status

- Fixed the issue where Safari browser does not trigger interaction when switching between Chinese and English languages

## V1.3.0（2025-01-30）

- Added file upload function

   - This function does not support internet connectivity when in use.

   - Supported formats: TXT, PDF, JPG, JPEG, PNG  

   - Scope of Application: Domestic and overseas (Chinese-English compatible), PC and mobile terminals (adapted)

## V1.2.2（2025-01-13）

- Optimized the timeout fallback logic for crawler services, resolving issues such as failed paper parsing

## V1.2.1（2025-01-09）

- New convenient code display features: support for sticky code copy button, fixed-height display, download, zoom, and toggle view

- Add a fallback logic for intent recognition timeout

## V1.2.0（2025-01-07）

- New web page parsing and crawler: support for obtaining more web page information, such as arXiv papers

- Fixed the issue where only the currently selected generated content is shared during regeneration

- Optimize the display of dark theme for editing queries

## V1.1.2（2025-12-26）

- New query editing function: Supports users to re-edit and resend sent queries to improve chat efficiency

- Optimize the regenerate function: Optimize the logic for regenerating based on the latest round of query in multi-turn conversations

## V1.1.1（2025-12-24）

- Optimize the display of user prompts for rate limiting

- Added support for line break input on mobile devices

- One-click copy for newly added Xiaomi ID

## V1.1.0（2025-12-20）

- New context management capability: Token prediction and truncation strategies launched based on single-turn and historical multi-turn inputs

- Optimized Deep Thinking: Automatically determine whether to enable deep thinking based on user input

## V1.0.2（2025-12-19）

- Optimize the web search capability: automatic web connection judgment, optimized filtering and re-ranking strategies, and newly added intent recognition and moderation strategies

- Fix the review processing logic

## V1.0.1（2025-12-18）

- Adjust the platform's positioning copy and display logic to more accurately describe the product positioning and AI compliance

- Added code to support serial number display

## V1.0.0（2025-12-17）

{/* feishu-style:text-align:left */}
**Global Launch of Xiaomi MiMo Studio**

- Model experience: Support MiMo-V2-Flash model experience

- Dialog Management: Supports basic management capabilities such as conversation search, renaming, and deletion

- Safety Management and Control: Supports content security review at the conversation level and security risk control capabilities at the account level

- User Settings: support functions including account login and logout, operation communication, user and privacy agreements, as well as filing information of large models and products, etc.

- Display interaction: supports 2 theme colors, and is adapted for display in scenarios including mobile devices, global English, SVG, and code display preview

- Platform Interconnection: Supports the jump entry between Platform and Studio platforms as well as account status synchronization


--- DOCUMENT: Xiaomi MiMo Code ---
URL: https://mimo.mi.com/static/docs/updates/feature/mimo-code.md

# Xiaomi MiMo Code

### v0.1.13 · 2026-08-19

{/* feishu-style:text-align:left */}
**Theme: Fixing Operational Stability**

- Each TUI session now provides a scoped OpenAI-compatible interface on a random local port, allowing Skills and child processes to access conversation, speech synthesis, and speech recognition capabilities via a restricted token without exposing the Provider credentials.

- Exec can directly invoke the MCP tools that have been authorized for the current request; the read-only `ALL_TOOLS `directory and TypeScript transpilation capability are newly added to reduce tool search round trips, and it is compatible with Node Bundle.

- Added `MIMOCODE_CODEX_MODE`, which enables any model to utilize a unified set of GPT/Codex prompts and toolkits; GPT models under Codex OAuth are no longer restricted to a fixed 372K context, and will instead follow the actual context configuration of the model.

- Checkpoint supports independent shutdown and rollback to context compression; the removal of automatic approval has been changed to instance-level control, `--yolo `still respects explicit rejection rules, and permission wait timeout is also decoupled from skip-all.

- Further strengthened the isolation between the Node runtime and the child process environment, and fixed issues including MCP/Skill sequence, child Agent recursion, rollback upon compression failure, Anthropic images and Thinking, OpenAI Responses, MiMo-V2.5 routing, and deletion confirmation popups.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.13)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.12...v0.1.13)

### v0.1.12 · 2026-08-13

{/* feishu-style:text-align:left */}
**Topic: Credential Isolation and Unattended Execution**

- Provider Key, OAuth Refresh Token, MCP environment variables and request headers are no longer passed to child processes such as Bash, MCP, LSP, and PTY along with the inherited environment; embedders may instead use in-process interfaces to inject credentials.

- Added `--yolo `as a shorthand for `--dangerously-skip-permissions `in the CLI; the deletion operation will skip secondary confirmation only when all targets are confirmed to be located in the system temporary directory and not belong to the current project.

- Compose Next can be discovered and invoked by the model upon explicit user authorization, while retaining the constraints of worktree switching, specification overriding and context restoration.

- For GPT service overload incidents, a maximum of three automatic retries are supported; models including GPT, Claude and Kimi will no longer receive inapplicable scheduled Skill Search reminders.

- Release the extended waiting window for mirror synchronization and return a non-zero status upon failure, while supplementing the SDK types as well as contribution and security instructions.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.12)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.11...v0.1.12)

### v0.1.11 · 2026-08-10

{/* feishu-style:text-align:left */}
**Topic: Controllable Memory Writing and MCP Parameter Reliability**

- Added `memory. disable_write`: stops writing to checkpoint, `MEMORY. md`, `notes. md `and task progress, while retaining memory retrieval capability.

- Added `memory-search `built-in skill, which supports querying the SQLite session history library.

- Fixed the issue where OpenAI-compatible streaming parsing caused silent parameter loss for multi-parameter MCP tools.

- The MCP Client has been adjusted to a process-level single-instance architecture, and issues related to child Agent prompts, session diff, and TUI session list have been fixed.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.11)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.10...v0.1.11)

### v0.1.10 · 2026-08-05

{/* feishu-style:text-align:left */}
**Topic: Orchestrator, Multi-session and Context Reliability**

- The Orchestrator preferentially routes to existing sessions, and the liveness status of sub-sessions is now determined based on the last activity time, which reduces misjudgments and duplicate creation.

- Pre-compress the budget according to the model settings, follow the context upper limit configured by the user, and fix the issues of duplicate reconstruction and checkpoint recovery.

- Exec restores MCP distribution and real-time sub-call traces, and MCP tools are found to support request-level, token-aware gating.

- The Provider no longer sends empty messages, and has enhanced the compatibility of Anthropic reasoning, OpenAI streaming retry, and DeepSeek reasoning effort.

- TUI adds Vivid / Minimal visual modes, with built-in Playwright CLI and Grok Build skill packs.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.10)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.9...v0.1.10)

### v0.1.9 · 2026-07-24

{/* feishu-style:text-align:left */}
**Topic: MiMo-V2.5 Free Limited Switch and Plan Mode Fix**

- MiMo-V2.5 will be available for free trial until July 26,2026, and TUI has added a corresponding end-of-life prompt; usage with a self-provided API Key will not be affected.

- Improve the continuation of the Plan mode under unconfirmed responses to prevent the model from exiting the planning process prematurely.

- Optimized slash command discovery, GPT sub-Agent tool parsing, and OpenAI response header wait timeout.

- Workflow tools are now enabled on demand, and the Python toolchain rules are aligned with the default Ruff set.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.9)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.8...v0.1.9)

### v0.1.8 · 2026-07-23

{/* feishu-style:text-align:left */}
**Topics: Codex Login, GPT Support, and Compose Next**

- Added the Codex login option, removed the model filter, and officially added support for the GPT model.

- Added built-in `/compose-next `to gradually replace the old Compose Agent.

- The write scope of Dream / Distill is limited to memory and the`. mimocode `directory.

- Office Skills adds OS-specific CJK font guidelines and supports discovery of packaged runtimes via the `MIMO_* `environment variables.

- Fixed issues with empty user messages, global session remount timing, and checkpoint writer recovery boundaries.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.8)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.7...v0.1.8)

### v0.1.7 · 2026-07-20

{/* feishu-style:text-align:left */}
**Topic: Scripted Tool Orchestration and Skill Search**

- Added `tool_script`, which allows the model to orchestrate multiple tool calls in the QuickJS sandbox to reduce round trips.

- Added a BM25-based `skill_search`, and added skill packages in the categories of Data Analysis, product design, sales and learning.

- Added optional Session Handoff and Try-best detectors.

- Upgrade xAI/Grok Provider and support OAuth login.

- We have hardened the serialization of tool scripts, sandbox isolation and MCP distribution, and fixed session issues such as GPT reasoning-only steps.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.7)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.6...v0.1.7)

### v0.1.6 · 2026-07-15

{/* feishu-style:text-align:left */}
**Topic: Multi-skill Orchestration, Permission Bypass and Custom Provider**

- Slash Skills is enabled by default; when multiple skills are referenced in the same message, they will be automatically loaded and an orchestration plan will be generated.

- Added `/skip-permissions `for unattended tasks: normal permissions are automatically granted, and dangerous operations are automatically rejected after timeout.

- Added `/modalities `to enable configuration of image, audio, video and PDF capabilities for custom models in the TUI.

- Unified into a single System Message to improve compatibility with self-deployed backends such as vLLM; fixed the schema issue of Moonshot / Kimi tool calls.

- Orchestrator adds liveness tracking, fan-in aggregation and laggy monitoring; newly added `/rebuild `for manual context rebuilding.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.6)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.5...v0.1.6)

### v0.1.5 · 2026-07-07

{/* feishu-style:text-align:left */}
**Topic: Built-in Skills, Research Workflow and Windows Installer**

- It comes with 14 built-in skills across 14 domains for the first time, covering scenarios such as Office document processing, academic writing, information retrieval and research, front-end design, and skill creation.

- Added `deep-research `and `fact-check `workflows to support parallel research, reflection and citation verification.

- The installation, upgrade and uninstallation process for native Windows PowerShell is now officially available.

- Image pasting supports multimodal capability judgment; Edit is changed to exact match by default, and fuzzy match becomes optional.

- Experimental Orchestrator and Cron/Loop scheduling capabilities are provided for the first time.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.5)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.4...v0.1.5)

### v0.1.4 · 2026-06-29

{/* feishu-style:text-align:left */}
**Topic: Compose Workflow and Observability**

- Workflow TUI has been upgraded to a full-screen observable page that displays real-time phases, statuses, and sub-Agent conversations.

- Added a built-in `compose `Workflow, which formalizes the sequence of Brainstorm → Design → Implement → Verify → Review → Merge into an executable parallelizable process.

- Added dedicated system prompts for DeepSeek, GLM and MiniMax.

- Fixed the garbled output of Windows Chinese Shell, tool invocation in text mode, and long connection timeout issues.

- Adjust the MCP private network connection policy: the private network MCP Server configured by users will no longer be incorrectly intercepted by SSRF rules.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.4)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.3...v0.1.4)

### v0.1.3 · 2026-06-24

{/* feishu-style:text-align:left */}
**Topic: Security Hardening and Chinese Input Experience**

- Fixed 25 issues including XSS, authentication, rate limiting, and Webhook signature verification based on security audit results.

- Added SSRF protection for AI-initiated Web Fetch, including defenses against private network, cloud metadata address and IPv4-mapped IPv6 bypasses.

- Fixed the issues of residual symbols, missing characters, and incorrect auto-completion alignment caused by multi-line pasting after Chinese characters.

- The model selector pins MiMo Provider to the top and unifies the OAuth login entry.

- Add a tool compatibility layer to unify the behaviors of tools including Edit, Read, Write, Apply Patch, and Notebook.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.3)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.2...v0.1.3)

### v0.1.2 · 2026-06-23

{/* feishu-style:text-align:left */}
**Topic: Free Channels, Command Search and Notebook Editing**

- Redesigned the display of the free MiMo channel, and added a Token Plan prompt when rate limiting is triggered.

- The Command Palette supports keyword search in both English and Pinyin, eliminating the need for Chinese users to switch between input methods.

- Added `notebook-edit`, which supports the replacement, insertion and deletion of Jupyter cells.

- Plan mode switching no longer refreshes the tool list, and voice input supports custom ASR Providers.

- The installation package is connected to the Chinese mainland CDN, and the protection against directory traversal and working directory boundaries has been strengthened.

{/* feishu-style:text-align:left */}
[View Release ](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.2)· [Full Changelog](https://github.com/XiaomiMiMo/MiMo-Code/compare/v0.1.1...v0.1.2)

### v0.1.1 · 2026-06-15

{/* feishu-style:text-align:left */}
**Topic: First Maintenance Release**

- GitHub Release only labels the version number and does not provide independent and detailed change descriptions.

- This version follows up on the iterations concerning installation, terminal compatibility and basic stability after the initial public release.

{/* feishu-style:text-align:left */}
[View Release](https://github.com/XiaomiMiMo/MiMo-Code/releases/tag/v0.1.1)

### v0.1.0 · 2026-06-11

{/* feishu-style:text-align:left */}
**Topic: MiMo Code Officially Released for the First Time**

- We have released a native AI programming Agent for terminals, which supports integration with multiple Providers and MiMo models.

- It introduces persistent memory, long context management, multi-agent collaboration and the Compose workflow pattern.

- Provides a configuration system covering TUI, MCP, Skills, voice input, and project-level`. mimocode`.

- The project is open-sourced under the MIT License, and supports installation on macOS, Linux and Windows.

### Project Link

- [GitHub Releases](https://github.com/XiaomiMiMo/MiMo-Code/releases)

- [MiMo Code Official Website](https://mimo.xiaomi.com/zh/mimocode)

- [MiMo Code GitHub](https://github.com/XiaomiMiMo/MiMo-Code)