Skip to content

Minimax

Minimax is a Chinese AI company providing language models with an OpenAI-compatible API.

Configuration

PROVIDER=minimax
API_KEY=your-minimax-api-key
MODEL=MiniMax-M2.1
TEMPERATURE=0.3
TOP_P=1.0
MAX_TOKENS=1024

Key Parameters

Parameter Description Default
API_KEY Minimax API key (required) -
MODEL Model to use -
TEMPERATURE Randomness (0.0-1.0) 0.3
TOP_P Nucleus sampling 1.0
MAX_TOKENS Max response tokens 1024
TIMEOUT Request timeout (seconds) 60
BASE_URL Custom API endpoint https://api.minimaxi.com/v1
EXTRA_BODY Additional request parameters -

Features

  • ✅ Streaming responses
  • ✅ Function calling
  • ✅ MCP support
  • ✅ OpenAI-compatible API
  • ✅ Chinese language optimization
  • ✅ Multi-modal capabilities
  • ✅ Interleaved Thinking (reasoning)

Interleaved Thinking

MiniMax-M2.1 and newer models support Interleaved Thinking (交错思维链), which allows the model to reason before each tool use and make decisions based on tool results.

How It Works

When reasoning_split=True is enabled (default), the API returns reasoning content separately from the main response:

{
    "message": {
        "content": "The weather in Beijing is sunny...",
        "role": "assistant",
        "reasoning_details": [
            {
                "type": "reasoning.text",
                "text": "I need to check the weather for Beijing..."
            }
        ]
    }
}

Configuration

Interleaved Thinking is enabled by default for MiniMax provider. You can control it via:

Environment Variable: 

export YAI_MINIMAX_REASONING_SPLIT=true

Config File: 

[core]
MINIMAX_REASONING_SPLIT=true

EXTRA_BODY (advanced): 

EXTRA_BODY={"reasoning_split": true}

Note: When reasoning_split is explicitly set in EXTRA_BODY, it takes precedence over MINIMAX_REASONING_SPLIT.

Important Notes

  • Reasoning content is preserved in conversation history and sent back to the model in subsequent requests
  • This ensures the model maintains continuity in its thinking process across multiple tool calls
  • Reasoning content is displayed alongside the main response when SHOW_REASONING=true

Important Notes

  • Uses OpenAI-compatible client implementation
  • Strong performance on Chinese language tasks
  • Supports image understanding in compatible models
  • Requires registration on the Minimax platform
  • API documentation primarily available in Chinese
  • Pay-as-you-go billing based on token usage

Resources