MiniMax M3.1 Flash

Looking for MiniMax 3.1 Flash? The official model name is MiniMax M3.1 Flash. This independent guide helps you choose an access route, estimate API costs, and prepare a useful first task.

Start here: Open MiniMax Code · Estimate API costs · Copy an API example

Independent guide, unaffiliated with MiniMax. Facts and prices checked October 23, 2026. Access and billing depend on your account.

What is MiniMax M3.1 Flash?

MiniMax M3.1 Flash is a model for coding and tasks that involve planning and using tools. Its documented context window is up to 1 million tokens. It accepts text, images, and video and returns text. See the official model guide.

For a first evaluation, choose work you can check: a small bug with a reproduction case, a document with questions you can verify, or a screenshot with a clear analysis goal. A model's tool-use capability does not give a plain chat request access to your repository, browser, or terminal; the surrounding application supplies those tools.

Choose your starting point

API usage and a subscription are different billing routes. Check the resources associated with your key before sending requests. See account setup.

MiniMax M3.1 Flash pricing

The published API rates are $0.15 per million new input tokens, $0.015 per million cached input tokens, and $0.60 per million output tokens. Rates checked October 23, 2026; consult official pricing before budgeting.

Estimate the token cost of your workload

Use token counts, rather than word or character counts. These worked examples apply the listed rates; they are calculations, not measured model runs.

New input tokens = Total input tokens − Cached input tokens

Estimated cost in USD =
  (New input tokens × 0.15
   + Cached input tokens × 0.015
   + Output tokens × 0.60) / 1,000,000

Cached tokens are part of total input, not extra input. Do not count them twice. The caching documentation explains how to read usage and identify cache hits.

This estimates token charges only. Include billed reasoning in your output-token budget, and allow for retries, repeated conversation history, tool usage, and any other applicable charges. An agent task can make many requests. Subscription allowances and taxes are outside these examples.

How to use the MiniMax M3.1 Flash API

The following request uses MiniMax's OpenAI-compatible Chat Completions endpoint. It sends a short task and requests a bounded response. Adapted from the official endpoint reference; it has not been executed for this guide.

1. Get an active key from the MiniMax developer platform.
2. Set MINIMAX_API_KEY in your local shell or server environment.
3. Run the request locally or on your server. It can consume paid resources.

# Prerequisite: MINIMAX_API_KEY is set in this shell.
curl --fail-with-body --silent --show-error \
  'https://api.minimax.io/v1/chat/completions' \
  --header "Authorization: Bearer ${MINIMAX_API_KEY:?Set MINIMAX_API_KEY first}" \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "MiniMax-M3.1-Flash",
    "reasoning_effort": "medium",
    "max_completion_tokens": 4096,
    "messages": [
      {
        "role": "user",
        "content": "Write a JavaScript function that groups an array of objects by a supplied property. Explain how missing properties are handled and include three test cases."
      }
    ]
  }'

Read the final answer from choices[0].message.content. Inspect the response's usage and finish_reason; a length-limited response may require a larger output budget. Keep API keys out of public HTML, client-side scripts, and shared request logs.

Adjust reasoning for the task

The API supports low, medium, high, xhigh, and max effort. If omitted, effort defaults to max. Reasoning cannot be disabled for this model. Start with a setting appropriate to your task, then compare answer quality, latency, and billed usage. See the parameter details.

If your first request fails

These are MiniMax provider codes, not HTTP status codes. Use the returned error details and official error reference to diagnose the request.

MiniMax M3.1 Flash vs M3: compare the cost of your task

For a practical first comparison, hold the prompt, input files, tools, and success criteria constant. Compare the bill for the complete task, including failed attempts, rather than just one token rate.

This compares the currently displayed API rates, not subscription prices. M3 tiers and discounts can change; see the pricing table. A cheaper request does not establish better accuracy or lower total task cost.

What do the published benchmarks show?

MiniMax's launch announcement reports 73.7% on DeepSWE v1.1, 82% on CyberGym, and 61.7 points on OSWorld V2. These are provider-reported evaluations, not tests performed by this site. They cover different tasks and should not be combined into an overall score. Read the original announcement for context.

Evaluate it on work you can verify

Try the same small task with your current model and M3.1 Flash. Record the model ID, date, effort, tools, elapsed time, usage, and whether the result met your acceptance criteria.

• Coding: Give both models the same failing example. Check the proposed fix and whether it changes unrelated behavior.
• Document work: Ask for an answer grounded in supplied documents. Check every important assertion against the original material.
• Tool workflows: Define what counts as completion before starting. Inspect the resulting files or application state, not just the final chat message.

A useful first-task prompt:

Goal: Fix the reproduced issue in the supplied project.
Inputs: Repository, steps to reproduce, expected behavior, and a failing example.
Constraints: Preserve existing behavior outside this issue. Explain assumptions.
Acceptance: Demonstrate the fix using the reproduction case and relevant checks.
Deliver: Changed files, verification evidence, and any unresolved limitations.

This is a task template, not a recorded success story. Supply your actual files and verify the output before relying on it.

Frequently asked questions

Is MiniMax 3.1 Flash the same as MiniMax M3.1 Flash?

This guide uses the shortened search phrase to refer to MiniMax M3.1 Flash. Use the exact identifier MiniMax-M3.1-Flash in API requests; the current model list identifies the supported model.

Can I try it for free?

Check the current MiniMax Code access options for your account. Free access, promotions, and limits can change. Do not assume that an API call is free because an app offers a free tier.

Can I try the model on this page?

This page provides instructions, cost examples, and links to the provider. It does not send prompts to a model or host a chat service. To run a task, open MiniMax Code or use your own API access.

Can I download the model weights?

The October 19 announcement said weights would follow later. That statement alone does not establish today's download availability or license. Check MiniMax's current model documentation and follow an officially linked release before downloading. This guide does not verify a weight download.

Does this guide include an independent review?

It includes sourced specifications and transparent cost calculations. It does not include a hands-on model review or measured competitor ranking. Use the evaluation checklist above to assess your own workload.

Ready to try a real task?

Choose one bounded task, define a result you can check, and keep a record of usage and outcome.

Open MiniMax Code · Read the API documentation · Check current pricing