@amux.ai/adapter-minimax

MiniMax adapter for Amux - bidirectional LLM API adapter.

Features

• ✅ OpenAI-compatible API format
• ✅ Interleaved Thinking support (reasoning_details)
• ✅ Streaming support
• ✅ Function calling / Tool use
• ✅ System prompts
• ✅ JSON mode
• ✅ Temperature range validation (0.0, 1.0]

Installation

pnpm add @amux.ai/adapter-minimax

Usage

Basic Example

import { createBridge } from '@amux.ai/llm-bridge'
import { minimaxAdapter } from '@amux.ai/adapter-minimax'
import { openaiAdapter } from '@amux.ai/adapter-openai'

const bridge = createBridge({
  inboundAdapter: openaiAdapter,
  outboundAdapter: minimaxAdapter,
  apiKey: process.env.MINIMAX_API_KEY!,
})

const response = await bridge.chat({
  messages: [
    { role: 'user', content: 'Hello!' },
  ],
})

console.log(response.choices[0]?.message.content)

Streaming with Interleaved Thinking

import { createBridge } from '@amux.ai/llm-bridge'
import { minimaxAdapter } from '@amux.ai/adapter-minimax'
import { openaiAdapter } from '@amux.ai/adapter-openai'

const bridge = createBridge({
  inboundAdapter: openaiAdapter,
  outboundAdapter: minimaxAdapter,
  apiKey: process.env.MINIMAX_API_KEY!,
})

// Enable reasoning split to get reasoning_details
const ir = openaiAdapter.inbound.parseRequest({
  model: 'MiniMax-M2.1',
  messages: [
    { role: 'user', content: 'Solve this complex problem...' },
  ],
  stream: true,
})

// Add reasoning_split extension
ir.extensions = {
  minimax: {
    reasoning_split: true,
  },
}

const stream = await bridge.chatStream(ir)

for await (const event of stream) {
  if (event.type === 'reasoning') {
    console.log('Thinking:', event.reasoning?.delta)
  } else if (event.type === 'content') {
    console.log('Content:', event.content?.delta)
  }
}

Supported Models

• MiniMax-M2.1 - Strong multilingual programming capabilities (output speed ~60 tps)
• MiniMax-M2.1-lightning - Lightning version: faster and more agile (output speed ~100 tps)
• MiniMax-M2 - Designed for efficient coding and Agent workflows

Capabilities

| Feature | Supported | |---------|-----------| | Streaming | ✅ | | Tools | ✅ | | Vision | ❌ | | Multimodal | ❌ | | System Prompt | ✅ | | Tool Choice | ✅ | | Reasoning | ✅ | | Web Search | ❌ | | JSON Mode | ✅ | | Logprobs | ❌ | | Seed | ❌ |

API Endpoint

• Base URL: https://api.minimaxi.com
• Chat Path: /v1/chat/completions
• Models Path: /v1/models

Special Features

Interleaved Thinking

MiniMax M2.1 supports Interleaved Thinking through the reasoning_details field. To enable it:

1. Set reasoning_split: true in the request (via extensions)
2. The reasoning content will be separated into reasoning_details array
3. Each detail has type: 'thinking' and text fields

Temperature Range

MiniMax has a strict temperature range of (0.0, 1.0]:

• Minimum: 0.01 (values ≤ 0.0 are clamped to 0.01)
• Maximum: 1.0 (values > 1.0 are clamped to 1.0)
• Recommended: 1.0

Unsupported OpenAI Parameters

The following OpenAI parameters are not supported by MiniMax:

• presence_penalty
• frequency_penalty
• logit_bias
• logprobs
• seed
• n (only supports value 1)

Documentation

For more information about MiniMax models:

• MiniMax Text Generation Guide
• OpenAI API Compatibility
• MiniMax M2.1 Release

License

MIT