minimaxText Chat

MiniMax M3 API: long-context text model for documents and agent memory

Use MiniMax M3 for long-context document review, agent memory inspection, knowledge workflows, and production text automation.

InputAIReiter $0.60 per 1M tokensOutputAIReiter $2.40 per 1M tokensCache readAIReiter $0.12 per 1M tokens

Run with API

INPUT

FormJSONNode.jsPythonHTTP

OUTPUT

Example

PreviewJSON

A rate limit caps how many requests an API accepts from you in a given window. Once you exceed it, the server stops doing work for you and answers 429 Too Many Requests instead.

Handling 429

Read the Retry-After response header. When present it tells you exactly how long to wait, in seconds.
When it is absent, back off exponentially with jitter so retries from many clients do not line up.
Cap the number of retries, then surface the failure instead of looping forever.

async function withRetry(request, maxRetries = 4) {
  for (let attempt = 0; ; attempt++) {
    const response = await request();
    if (response.status !== 429 || attempt === maxRetries) return response;
    const retryAfter = Number(response.headers.get("retry-after"));
    const backoff = Number.isFinite(retryAfter) ? retryAfter * 1000 : 2 ** attempt * 500 + Math.random() * 250;
    await new Promise((resolve) => setTimeout(resolve, backoff));
  }
}

Treat the limit as a budget you plan around, not an error you retry your way out of: batch requests where you can, cache repeated reads, and spread bulk work over time.

{
  "model": "minimax-m3",
  "input": {
    "model": "minimax-m3",
    "max_tokens": 2048,
    "messages": [
      {
        "role": "user",
        "content": "Read this long project note and extract decisions, open questions, and owners."
      }
    ],
    "temperature": 0.7,
    "top_p": 1
  },
  "output": "A rate limit caps how many requests an API accepts from you in a given window. Once you exceed it, the server stops doing work for you and answers **429 Too Many Requests** instead.\n\n## Handling 429\n\n1. Read the `Retry-After` response header. When present it tells you exactly how long to wait, in seconds.\n2. When it is absent, back off exponentially with jitter so retries from many clients do not line up.\n3. Cap the number of retries, then surface the failure instead of looping forever.\n\n```js\nasync function withRetry(request, maxRetries = 4) {\n  for (let attempt = 0; ; attempt++) {\n    const response = await request();\n    if (response.status !== 429 || attempt === maxRetries) return response;\n    const retryAfter = Number(response.headers.get(\"retry-after\"));\n    const backoff = Number.isFinite(retryAfter) ? retryAfter * 1000 : 2 ** attempt * 500 + Math.random() * 250;\n    await new Promise((resolve) => setTimeout(resolve, backoff));\n  }\n}\n```\n\nTreat the limit as a budget you plan around, not an error you retry your way out of: batch requests where you can, cache repeated reads, and spread bulk work over time.",
  "metrics": {
    "input_tokens": 26,
    "output_tokens": 214,
    "generated_in_seconds": 4.1
  },
  "example": true
}

Generated in

4.1 seconds

Input tokens

Output tokens

214

Tokens per second

52.20 tokens / second

Time to first token

Model details

Use the same model key in Playground, API requests, and internal workflows.

Model ID: minimax-m3
Provider: minimax
Protocol: Anthropic Messages
Context window: 1,000,000 tokens
Max output: -
Input tokens: 60 credits / 1M tokens
Output tokens: 240 credits / 1M tokens
Cache read: 12 credits / 1M tokens
Cache write: -

A long-context route for documents and agent memory

MiniMax M3 is useful when a workflow must keep more evidence in one request: policy sets, research packs, meeting history, or long agent memory.

Should you choose MiniMax M3?

Position it as a practical long-context model for document-heavy teams and agent workflows that need continuity.

Choose it when

You need long document review, memory inspection, knowledge-base synthesis, or a cheaper long-context route before escalating to a flagship model.

Use another model when

The task is a small chat turn, simple extraction, or high-risk code reasoning that needs a more specialized coding model.

Public API protocol

Call POST https://aireiter.com/api/v1/messages with model "minimax-m3". Streaming is supported through the same Messages-compatible endpoint.

Token and cache usage

Pricing is based on input, cache-read, and output usage. For long prompts, cache-read fields matter because reused context can materially change cost.

MiniMax M3 production workloads

Best for long inputs where continuity and summarization quality matter.

Document packs

Review policies, reports, transcripts, and research notes while keeping cross-document references visible.

Agent memory

Inspect long histories, tool calls, and state updates to explain why an agent made a decision.

Knowledge workflows

Turn long internal material into summaries, briefs, requirements, and action lists.

Cost-aware long context

Evaluate whether a practical long-context model is enough before using a higher-cost flagship route.

How MiniMax M3 fits your model stack

Do not route every request to the newest model. Pick the cheapest model that still passes your quality bar, then reserve deeper models for failures or high-risk tasks.

For fast batches

Use Doubao or DeepSeek V4 Flash for small fast tasks; MiniMax M3 is for context-heavy inputs.

For deeper reasoning

Use GLM 5.2 or DeepSeek V4 Pro when reasoning depth is more important than context size.

For long context

Compare MiniMax M3 with Kimi K2.7 Code when the workload mixes documents, code, and agent traces.

For production rollout

Track cache-read usage on repeated long prefixes; this is where long-context workflows can become more economical.

MiniMax M3 API questions

Questions developers usually check before moving a text model from playground testing to production API traffic.

/ 01

What model ID should I send for MiniMax M3?

Use "minimax-m3" in the API request body. The internal DB key is only used by AIReiter routing.

/ 02

Which endpoint should MiniMax M3 use?

Use POST https://aireiter.com/api/v1/messages for public API calls. Keep x-api-key / Authorization authentication consistent with your AIReiter API key setup.

/ 03

Does MiniMax M3 support streaming?

Yes. Send stream=true and read server-sent events until the message completes. Test non-streaming first when debugging authentication or model ID issues.

/ 04

How do I confirm token and cache billing for MiniMax M3?

Check the usage object returned by the API. Input, output, and cache-read token fields are the source of truth for settlement; a repeated prompt alone does not prove a cache hit.

/ 05

Should I always set max_tokens for MiniMax M3?

Set max_tokens according to the expected summary length. Long input does not always require huge output, but review tasks need enough room for findings and evidence.