MiniMax-M3.1-Flash
minimax/minimax-m3.1-flash
Back to model catalog

MiniMax · Model guide

MiniMax-M3.1-Flash

Compare
minimax/minimax-m3.1-flashReleased Oct 19, 2026

The Flash model in MiniMax's M3.1 series: a natively multimodal agent model with a 1M-token context window and up to 512K output tokens, taking text, image and video input. Thinking is always on, and its depth is set with five effort levels from low to max (max by default). MiniMax positions it for agentic tasks, coding, computer use and cybersecurity, with online inference nearly three times faster than its predecessor.

Modalities
TextImageVideoText
Context
1M
Max output
524K
Price / 1M tokens
$ 0.15in$ 0.60out
StreamingTool callingJSON outputImage inputThinking mode

What this model is good at

What the vendor positions it for, and which of our models to reach for instead.

VendorMiniMax presents M3.1 Flash as a generational leap in agentic tasks, coding, computer use and cybersecurity, with stronger interaction with real environments, tool use and execution of long task sequences.source

Best for
  • Coding agents · The vendor reports DeepSWE v1.1 at 73.7 (mini-SWE-agent) and Terminal-Bench 4.0 at 38.1 (Codex harness).source

  • Computer use and workflow automation · The vendor reports OSWorld 2.0 at 61.7, Automation-Bench v1.0.6 at 62.8 and Agents' Last Exam at 29.6.source

  • Agents that read images and video · Image and video input are native to the model, alongside text.source

  • Fast agent loops · MiniMax says online inference is nearly three times faster than its predecessor, at over 200 tokens per second in everyday use.(vendor claim)source

Neighbouring models
  • minimax/minimax-m3predecessor —Choose M3 when you need to switch thinking off: M3 skips thinking when it is disabled, while M3.1 Flash always thinks and rejects a disabled thinking setting.source
Compare them side by side →

Pricing and billing

Billed per token. Cached input is charged at the cache-read rate.

Price / 1M tokensYou pay
Input$ 0.15
Output$ 0.60
Cache read$ 0.015
Cache write$ 0.15
Estimate a request
$ 0.0027
Estimated cost per request
10.0K × $ 0.15
+ 2.0K × $ 0.60
≈ $ 2.70 per 1,000 requests

Capabilities and limits

What CrossModel guarantees across every route this model can take right now.

Context window
1.0M tokens
Max output
524.3K tokens
Input / output modalities
Text + Image + Video → Text
Streaming
Supported
Tool calling
Supported
Structured output (JSON)
Supported
Image input
Supported
Thinking mode
Supported· always on
Reasoning effort
low · medium · high · xhigh · max
Available endpoints
/v1/chat/completions · /v1/responses · /v1/messages

Published benchmarks

Scores the vendor reported, with the evaluation setup each one came from.

Agentic coding
  • Terminal-Bench 4.0
    pass_rate·MiniMax · 2026-10-19Codex 0.149.1setup not fully disclosed

    厂商自测。模型卡注明使用 Codex 0.149.1 框架;目录里 DeepSeek V4.1 Flash 同一身份的成绩用的是 Claude Code 框架,框架不同,直接比较时留意。未披露推理档位与采样设置。

    38.1%
Agentic work
  • Agents' Last Exam
    score·MiniMax · 2026-10-19setup not fully disclosed

    模型卡注明采用官方 Overall 设置,全集成绩取自 ALE 官方榜单;发布博客正文同为 29.6。未披露推理档位与采样设置。

    29.6%
  • AutomationBench v1.0.6
    score·MiniMax · 2026-10-19setup not fully disclosed

    厂商自测。模型卡写作 Automation-Bench v1.0.6,在公开发布的 600 道题上测试;未披露推理档位与采样设置。

    62.8%
Coding
  • DeepSWE 1.1
    resolved_rate·MiniMax · 2026-10-19mini-SWE-agent 2.4.5setup not fully disclosed

    厂商自测。模型卡注明使用 mini-SWE-agent 2.4.5、512K 上下文;发布博客正文同样给出 73.7%。未披露推理档位与采样设置。

    73.7%
  • NL2Repo
    score·MiniMax · 2026-10-19Claude Code 2.1.233setup not fully disclosed

    厂商自测。模型卡写作 NL2Repo-Bench,注明使用 Claude Code 2.1.233、512K 上下文,并修正了两道不一致的题目(成绩基于修正后的题集),与其它厂商的成绩不完全可比;只给出百分数,保守记为 score。

    61.2%
Computer use
  • OSWorld 2.0
    success_rate·MiniMax · 2026-10-19setup not fully disclosed

    厂商自测。模型卡注明使用 OSWorld 2.0 的 0808 Full 版本(不是离线子集);发布博客正文写作 OSWorld V2,同为 61.7。同表中 Claude Opus 5 的 70.6 与目录该身份的既有成绩一致。未披露推理档位、步数上限与采样设置。

    61.7%
Cybersecurity
  • CyberGym
    score·MiniMax · 2026-10-19Claude Code 2.1.263setup not fully disclosed

    厂商自测。模型卡注明使用 Claude Code 2.1.263、1M 上下文;发布博客正文取整写作 82%。同表中 GLM-5.3 的 84.5 与目录该身份的既有成绩一致。未披露推理档位与采样设置。

    82.2%
Ml engineering
  • PostTrainBench 1.1
    score·MiniMax · 2026-10-19Claude Code 2.1.251setup not fully disclosed

    厂商自测。模型卡注明使用 Claude Code 2.1.251、512K 上下文,其余按官方设置;未披露推理档位与采样设置。

    39.3%
Multimodal reasoning
  • BabyVision (with tools)
    score·MiniMax · 2026-10-19Claude Code 2.1.285setup not fully disclosed

    厂商自测。模型卡注明使用 Claude Code 2.1.285,图片最短边不小于 1,500 像素;未披露具体工具集合与推理档位。

    91%

Vendor-reported numbers, not CrossModel measurements. Scores are only comparable when the benchmark version, metric and evaluation setup match, so nothing here is averaged or ranked.

Use it in your tools

Point the base URL at CrossModel and paste this model ID — every tool below has a setup guide.

See all integration guides →

Frequently asked questions

What is MiniMax-M3.1-Flash?
MiniMax-M3.1-Flash is available on CrossModel as minimax/minimax-m3.1-flash. It has a 1M-token context window and can return up to 524K tokens per request.
How much does MiniMax-M3.1-Flash cost?
Currently $ 0.15 per 1M input tokens and $ 0.60 per 1M output tokens. Cached input is billed at $ 0.015 per 1M.
Does MiniMax-M3.1-Flash support tool calling and structured output?
Tool calling: Supported · Structured output (JSON): Supported · Image input: Supported · Streaming: Supported · Thinking mode: Supported
Which endpoint do I call?
For text models, OpenAI Chat Completions (/v1/chat/completions), OpenAI Responses (/v1/responses) and Anthropic Messages (/v1/messages) all work — pick whichever your SDK already speaks.
Can I try it without writing code?
Yes. Sign in and open the Playground on this page to chat with the model directly — no API key needed. Usage is billed from your wallet at the normal rate.

Chat first, integrate later

Chat first, then wire it in once you like the answers.