Update, October 19, 2026: This article has been updated after MiniMax released M3.1 Flash. An earlier version said MiniMax had announced the model was coming.

MiniMax has released MiniMax-M3.1-Flash. A post on MiniMax's blog dated October 19 says the company is officially rolling out the model, which is available through the MiniMax API, M Plan and MiniMax Code. Earlier on October 19, at 17:21 Beijing time, MiniMax's official account on X had posted "MiniMax-M3.1-Flash is coming."

MiniMax's pricing page lists M3.1 Flash at $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million tokens for prompt caching reads, with a single price tier. MiniMax's China pricing page lists 1 yuan for input, 4 yuan for output and 0.1 yuan for cache reads per million tokens.

Space Bunny

MiniMax's blog says Space Bunny, "which processed more than 200 trillion tokens while operating anonymously, was an early version of MiniMax M3.1 Flash."

What MiniMax says about performance and cost

MiniMax says M3.1 Flash makes online inference nearly three times faster and that, for agentic tasks, its overall usage cost is 70% lower than its predecessor's. The blog says output speeds can reach more than 200 tokens per second, with a median above 170.

MiniMax reports these benchmark results:

  • DeepSWE v1.1: 73.7%
  • OSWorld V2: 61.7
  • Agents' Last Exam (ALE): 29.6
  • BankerToolBench: 76.2
  • CyberGym: 82%

MiniMax says M3.1 Flash ranks first among Chinese models on OSWorld V2 and ALE. For AutomationBench-AA, the blog doesn't give a score and says the model ranks among the top performers.

Weights and other availability

MiniMax's release post on X says the weights will arrive on October 23. Its blog says only that they will be released later. The blog also says a Pro version will arrive later.

On OpenRouter, the model is listed as minimax/minimax-m3.1-flash:free, with a 1 million token context window.

What MiniMax's docs say about M3.1 Flash

MiniMax's model list describes MiniMax-M3.1-Flash as a "Frontier multimodal coding model with a 1M context window and tunable thinking depth" and lists it first among its language models, ahead of MiniMax-M3.

MiniMax's API reference for its Anthropic-compatible endpoint says MiniMax-M3.1-Flash natively supports text, image and video input and "always thinks". A request that tries to turn thinking off returns an HTTP 400 error. Thinking depth is set with the effort field under output_config, which accepts low, medium, high, xhigh and max, and defaults to max. For max tokens, the reference recommends 131,072 (128K) and allows up to 524,288 (512K), the same values it gives for MiniMax-M3.

MiniMax had offered a preview version before the release. On September 27, Beijing time, MiniMax's agent account said M3.1-Flash-Preview had debuted on MiniMax Code, and at 01:06 Beijing time on September 28, MiniMax's official account said M3.1 Flash Preview was live on the Token Plan. MiniMax's M Plan overview now says MiniMax-M3.1-Flash-Preview has been retired and that requests using the old name are served by MiniMax-M3.1-Flash. M3.1 Flash is the text model on all three M Plan tiers, Go, Explore and Build.

The M2.5 and M2.7 retirement schedule

According to a retirement notice in MiniMax's model list, the M2.5 and M2.7 model families are no longer available to new users from October 19, 2026, and retire on November 18, 2026. MiniMax's API docs list MiniMax-M2.7, M2.7-highspeed, M2.5 and M2.5-highspeed among its legacy models.

For M Plan and Token Plan subscribers, the notice says that starting October 19, all requests to M2.5 and M2.7 models under those plans are automatically routed to M3.1 Flash. MiniMax says the existing model names keep working for backward compatibility, so requests don't need new model names, and plan usage allowances stay the same.

Existing pay-as-you-go API users can keep calling these models until November 18. After that, MiniMax says, API requests to them will no longer work. MiniMax asks users to check whether their applications rely on these models and to migrate before the retirement date.