ContextQMD
Back to Promptfoo

MiniMax Provider

site/docs/providers/minimax.md

0.121.132.7 KB
Original Source

MiniMax

MiniMax provides an OpenAI-compatible API for their language models. The MiniMax provider follows the OpenAI provider chat configuration pattern, with the MiniMax-specific parameter differences described below.

Setup

  1. Get an API key from the MiniMax Platform
  2. Set MINIMAX_API_KEY environment variable or specify apiKey in your config

Configuration

Basic configuration example:

yaml
providers:
  - id: minimax:MiniMax-M2.7
    config:
      temperature: 0.7
      max_completion_tokens: 2048
      apiKey: YOUR_MINIMAX_API_KEY

  - id: minimax:MiniMax-M2.7-highspeed
    config:
      max_completion_tokens: 2048

Configuration Options

  • temperature - Range (0.0, 1.0], cannot be 0
  • max_completion_tokens - Maximum completion tokens; the OpenAI-compatible API currently allows up to 2048. Legacy max_tokens config is translated to this field for compatibility.
  • apiBaseUrl - Optional custom MiniMax-compatible proxy endpoint
  • top_p
  • tools and tool_choice - Use these for tool calling. MiniMax rejects the deprecated function_call parameter.

When MiniMax reports prompt-cache reads, promptfoo calculates cost using the returned cached token count and the model's cache-read rate.

Available Models

MiniMax-M2.7 (Default)

  • Latest flagship model with enhanced reasoning and coding
  • 204,800 token context window
  • Input: $0.06/1M (cache hit), $0.3/1M (cache miss)
  • Output: $1.2/1M

MiniMax-M2.7-highspeed

  • High-speed version of M2.7 for low-latency scenarios
  • 204,800 token context window
  • Input: $0.06/1M (cache hit), $0.6/1M (cache miss)
  • Output: $2.4/1M

MiniMax-M2.5

  • Peak performance model with ultimate value
  • 204,800 token context window
  • Input: $0.03/1M (cache hit), $0.3/1M (cache miss)
  • Output: $1.2/1M

MiniMax-M2.5-highspeed

  • Same performance, faster and more agile
  • 204,800 token context window
  • Input: $0.03/1M (cache hit), $0.6/1M (cache miss)
  • Output: $2.4/1M

Example Usage

Here's an example comparing MiniMax with OpenAI:

yaml
providers:
  - id: minimax:MiniMax-M2.7
    config:
      temperature: 0.7
      max_completion_tokens: 2048
  - id: openai:gpt-4o
    config:
      temperature: 0.7
      max_tokens: 4000

prompts:
  - 'Answer the following question: {{question}}'

tests:
  - vars:
      question: 'What is the capital of France?'
    assert:
      - type: contains
        value: 'Paris'

API Documentation