OpenAI SDK

To meet developers’ needs for the OpenAI API ecosystem, our API now supports the OpenAI API format. With simple configuration, you can integrate MiniMax capabilities into the OpenAI API ecosystem.

Quick Start

1. Install OpenAI SDK

pip install openai

npm install openai

2. Configure Environment Variables

export OPENAI_BASE_URL=https://api.minimax.io/v1
export OPENAI_API_KEY=${YOUR_API_KEY}

3. Call API

Python

from openai import OpenAI

client = OpenAI()

response = client.chat.completions.create(
    model="MiniMax-M3",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hi, how are you?"},
    ],
    # Set reasoning_split=True to separate thinking content into reasoning_details field
    extra_body={"reasoning_split": True},
)

print(f"Thinking:\n{response.choices[0].message.reasoning_details[0]['text']}\n")
print(f"Text:\n{response.choices[0].message.content}\n")

4. Important Note

In multi-turn function call conversations, the complete model response (i.e., the assistant message) must be append to the conversation history to maintain the continuity of the reasoning chain.

Append the full response_message object (including the tool_calls field) to the message history
- For native OpenAI API with MiniMax-M3 MiniMax-M2.7 MiniMax-M2.7-highspeed MiniMax-M2.5 MiniMax-M2.5-highspeed MiniMax-M2.1 MiniMax-M2.1-highspeed MiniMax-M2 models, the content field will contain <think> tag content, which must be preserved completely
- In the Interleaved Thinking compatible format, by enabling the additional parameter (reasoning_split=True), the model’s thinking content is provided separately via the reasoning_details field, which must also be preserved completely

Supported Models

When using the OpenAI SDK, the following MiniMax models are supported:

Model Name	Context Window	Description
MiniMax-M3	1,000,000	Latest M-series language model for agentic reasoning, tool use, coding, and long-context tasks
MiniMax-M2.7	204,800	Beginning the journey of recursive self-improvement (output speed approximately 60 tps)
MiniMax-M2.7-highspeed	204,800	M2.7 Highspeed: Same performance, faster and more agile (output speed approximately 100 tps)
MiniMax-M2.5	204,800	Peak Performance. Ultimate Value. Master the Complex (output speed approximately 60 tps)
MiniMax-M2.5-highspeed	204,800	M2.5 highspeed: Same performance, faster and more agile (output speed approximately 100 tps)
MiniMax-M2.1	204,800	Powerful Multi-Language Programming Capabilities with Comprehensively Enhanced Programming Experience (output speed approximately 60 tps)
MiniMax-M2.1-highspeed	204,800	Faster and More Agile (output speed approximately 100 tps)
MiniMax-M2	204,800	Agentic capabilities, Advanced reasoning

For details on how tps (Tokens Per Second) is calculated, please refer to FAQ > About APIs.

For more model information, please refer to the standard MiniMax API documentation.

Multimodal Input

OpenAI-compatible Chat Completions support text, image, and video input for MiniMax-M3. Use image_url content parts for images and video_url content parts for videos. The detail field accepts low, default, or high and defaults to default; max_long_side_pixel can be used to control the longest side. Images support JPEG, PNG, GIF, and WEBP. Videos support MP4, AVI, MOV, and MKV; fps defaults to 1 and accepts values from 0.2 to 5. URL or base64 videos can be up to 50 MB, images can be up to 10 MB, and the request body can be up to 64 MB. For larger videos, upload through the Files API and pass mm_file://{file_id}; Files API videos can be up to 512 MB. Image token usage depends on image size and content. Use this as a rough single-image heuristic; check response usage or token counting where available for exact usage:

`detail`	Rough single-image token usage
`low`	Usually a few hundred tokens, up to ~600
`default`	Often ~1k-3k tokens, up to ~5k
`high`	Often several thousand tokens, up to ~15k+

Python

response = client.chat.completions.create(
    model="MiniMax-M3",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Summarize what is happening here."},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://example.com/image.png",
                        "detail": "default",
                    },
                },
                {
                    "type": "video_url",
                    "video_url": {
                        "url": "mm_file://file_id",
                        "detail": "default",
                    },
                },
            ],
        }
    ],
)

MiniMax-M3 Request Parameters

MiniMax-M3 supports these additional Chat Completions parameters through the OpenAI-compatible API:

Parameter	Description
`thinking`	Controls MiniMax-M3 thinking. `type` can be `disabled` or `adaptive`; when omitted, thinking is on by default. For M2.x models, thinking cannot be disabled.
`stream_options.include_usage`	When streaming, set to `true` to include token usage in the stream.
`max_tokens`	Legacy generation length limit.
`max_completion_tokens`	Generation length limit; use this field for new integrations.
`temperature`	Sampling temperature. Range `[0, 2]`, default `1`.
`top_p`	Nucleus sampling. Range `[0, 1]`. Default `0.95` for `MiniMax-M3` and `0.9` for M2.x models.
`tools`	Function tool definitions.
`reasoning_split`	Output-format switch. When enabled, separates thinking content into `reasoning_content` and `reasoning_details`.
`service_tier`	Request admission tier. Supported values are `standard` and `priority`; if omitted, requests use `standard`. The `priority` price is 1.5 times the `standard` price and ensures priority admission so the request is processed ahead of other requests, leading to faster responses and fewer failures.

Thinking Control

For MiniMax-M3, the thinking parameter controls whether the model can emit thinking content.

If thinking is omitted, thinking is on by default and the response includes thinking content.
Set thinking: {"type": "adaptive"} to explicitly keep thinking on. For MiniMax-M3, adaptive is equivalent to thinking on.
Set thinking: {"type": "disabled"} to skip thinking and answer directly.
For M2.x models, thinking cannot be disabled; thinking: {"type": "disabled"} is accepted but thinking remains on.

reasoning_split does not enable or disable thinking. It only controls how thinking content is returned: when true, thinking is exposed through reasoning_content and reasoning_details; when false, native Chat Completions responses keep thinking inside the content field with <think>...</think> tags.

Python

response = client.chat.completions.create(
    model="MiniMax-M3",
    messages=[{"role": "user", "content": "Hi, how are you?"}],
    extra_body={
        "thinking": {"type": "adaptive"},
    },
)

Examples

Streaming Response

Python

from openai import OpenAI

client = OpenAI()

print("Starting stream response...\n")
print("=" * 60)
print("Thinking Process:")
print("=" * 60)

stream = client.chat.completions.create(
    model="MiniMax-M3",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hi, how are you?"},
    ],
    # Set reasoning_split=True to separate thinking content into reasoning_details field
    extra_body={"reasoning_split": True},
    stream=True,
)

reasoning_buffer = ""
text_buffer = ""

for chunk in stream:
    if (
        hasattr(chunk.choices[0].delta, "reasoning_details")
        and chunk.choices[0].delta.reasoning_details
    ):
        for detail in chunk.choices[0].delta.reasoning_details:
            if "text" in detail:
                reasoning_text = detail["text"]
                new_reasoning = reasoning_text[len(reasoning_buffer) :]
                if new_reasoning:
                    print(new_reasoning, end="", flush=True)
                    reasoning_buffer = reasoning_text

    if chunk.choices[0].delta.content:
        content_text = chunk.choices[0].delta.content
        new_text = content_text[len(text_buffer) :] if text_buffer else content_text
        if new_text:
            print(new_text, end="", flush=True)
            text_buffer = content_text

print("\n" + "=" * 60)
print("Response Content:")
print("=" * 60)
print(f"{text_buffer}\n")

Tool Use & Interleaved Thinking

Learn how to use M3 Tool Use and Interleaved Thinking capabilities with OpenAI SDK, please refer to the following documentation.

Tool Use & Interleaved Thinking

Learn how to leverage MiniMax-M3 tool calling and interleaved thinking capabilities to enhance performance in complex tasks.

Important Notes

The temperature parameter range is [0, 2], recommended value: 1.0, values outside this range will return an error
Some OpenAI parameters (such as presence_penalty, frequency_penalty, logit_bias, etc.) will be ignored
Image and video inputs are supported by MiniMax-M3 through OpenAI-compatible message content parts; audio input is not currently supported
The n parameter only supports value 1
The deprecated function_call is not supported, please use the tools parameter

Using the API

Text

Speech

Video

Image

Music

File

Models

Quick Start

1. Install OpenAI SDK

2. Configure Environment Variables

3. Call API

4. Important Note

Supported Models

Multimodal Input

MiniMax-M3 Request Parameters

Thinking Control

Examples

Streaming Response

Tool Use & Interleaved Thinking

Tool Use & Interleaved Thinking

Important Notes

​Quick Start

​1. Install OpenAI SDK

​2. Configure Environment Variables

​3. Call API

​4. Important Note

​Supported Models

​Multimodal Input

​MiniMax-M3 Request Parameters

​Thinking Control

​Examples

​Streaming Response

​Tool Use & Interleaved Thinking

Tool Use & Interleaved Thinking

​Important Notes

Quick Start

1. Install OpenAI SDK

2. Configure Environment Variables

3. Call API

4. Important Note

Supported Models

Multimodal Input

MiniMax-M3 Request Parameters

Thinking Control

Examples

Streaming Response

Tool Use & Interleaved Thinking

Important Notes