MiniMax-M3

MiniMax-M3 is MiniMax’s M-series model for coding, agentic workflows, long-context tasks, and multimodal understanding. Through AnyFast, call it with the OpenAI-compatible POST /v1/chat/completions endpoint.

Key capabilities

1M-token context — Handle long documents, repositories, and multi-turn conversations.
Coding and agents — Supports tool calls and multi-step reasoning workflows.
Native multimodality — Accepts text, image, and video content in messages.
Adaptive thinking — Thinking is enabled by default for M3; use thinking to control it.
Reasoning output control — Use reasoning_split to separate reasoning into dedicated response fields.
Streaming — Receive incremental output with Server-Sent Events by setting stream to true.
Priority service — Choose service_tier: "priority" when your account has priority access.

Model ID

Use the official model ID MiniMax-M3 in the request body.

Quick example

cURL

curl --request POST \
  --url https://www.anyfast.ai/v1/chat/completions \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '
{
  "model": "MiniMax-M3",
  "messages": [
    {
      "role": "user",
      "content": "Explain how a compiler works in simple terms."
    }
  ],
  "thinking": {
    "type": "adaptive"
  },
  "max_completion_tokens": 2048,
  "stream": false
}
'

Multimodal input

Put text, image, or video parts in a message’s content array. Use a publicly accessible URL for image_url or video_url.

{
  "model": "MiniMax-M3",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "Describe this image." },
        {
          "type": "image_url",
          "image_url": { "url": "https://example.com/image.jpg" }
        }
      ]
    }
  ]
}

Thinking and reasoning

When thinking is omitted, MiniMax-M3 uses adaptive thinking by default. Set thinking.type to adaptive to enable adaptive thinking or disabled to skip thinking. The reasoning_split flag changes the response format; it does not turn thinking on or off. For M2.x models, thinking cannot be disabled. This page documents M3; use the model-specific page when calling another MiniMax model.

Streaming

Set stream to true to receive Server-Sent Events. Streaming responses use the OpenAI-compatible chat.completion.chunk object type.

cURL

curl --request POST \
  --url https://www.anyfast.ai/v1/chat/completions \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "MiniMax-M3",
    "messages": [{"role": "user", "content": "Write a short haiku about rain."}],
    "stream": true
  }'

Token limits

For MiniMax-M3, the recommended max_completion_tokens value is 131072 and the maximum is 524288. The official API also accepts the legacy max_tokens field, but it is deprecated; use max_completion_tokens for new integrations.

API Reference

View the interactive API reference for MiniMax-M3.

Getting started

Model Introduction

Resources

Key capabilities

Model ID

Quick example

Multimodal input

Thinking and reasoning

Streaming

Token limits

API Reference

​Key capabilities

​Model ID

​Quick example

​Multimodal input

​Thinking and reasoning

​Streaming

​Token limits

API Reference

Key capabilities

Model ID

Quick example

Multimodal input

Thinking and reasoning

Streaming

Token limits