Chinese AI company MiniMax released M3.1 Flash on October 19, 2026, through its API, its M Plan subscription and the MiniMax Code desktop app. MiniMax says the model runs nearly three times faster than M3 and cuts the cost of agentic tasks by 70%. It also confirmed that Space Bunny, an anonymous model that topped usage charts on OpenRouter and OpenCode before the launch, was an early version of M3.1 Flash, and said on X that the weights will be released on October 23.

Key Facts

Details
Release October 19, 2026
Where to use it MiniMax API (model ID MiniMax-M3.1-Flash), M Plan and MiniMax Code
International API price $0.15 input, $0.60 output and $0.015 cached input per million tokens, flat rate with no long-context surcharge (platform.minimax.io, billed in USD)
China API price ¥1.00 input, ¥4.00 output and ¥0.10 cache read per million tokens (platform.minimaxi.com, billed in RMB)
Compared with M3 M3's standard tier costs $0.30 input and $1.20 output per million tokens up to 512K input (¥2.10 and ¥8.40 in China), so M3.1 Flash's per-token price is half in USD and slightly less than half in RMB
Speed Over 200 output tokens per second in everyday use, with a median of over 170, according to MiniMax
Context and inputs 1M-token context; text, image and video input (MiniMax API docs)
Reasoning Always on. reasoning_effort accepts low, medium, high, xhigh or max, with max as the default; requests that try to switch thinking off return HTTP 400
Weights October 23, according to MiniMax's X post; the blog post says only that weights "will be released later". No licence announced

MiniMax reported these benchmark results for the model:

Benchmark M3.1 Flash
DeepSWE v1.1 73.7%
OSWorld V2 61.7
Agents' Last Exam (ALE) 29.6
BankerToolBench 76.2
CyberGym 82%
Automation-Bench 62% (X post)
BabyVision 91% (X post)

MiniMax says its OSWorld V2 and ALE scores rank first among Chinese models. All of these figures come from MiniMax.

Background

MiniMax released M3 on June 1, 2026, with a 1M-token context window and image and video input. Its weights are on Hugging Face under the minimax-community licence, and the model card lists about 428B total and 23B active parameters. M3.1 Flash is the first model in the 3.1 series. MiniMax says a Pro model "will also arrive later" and has not given a date.

Space Bunny ran anonymously before the launch. OpenRouter said it was the platform's top model for 10 straight days and processed 56.7 trillion tokens, the most of any stealth model OpenRouter has hosted. It left OpenRouter on October 5 and returned on October 19 under its real name, free for 72 hours. OpenCode said the model was its top model for 12 consecutive days and served 106 trillion tokens; it is free there for three days. MiniMax puts Space Bunny's total at more than 200 trillion tokens.

Reactions and Impact

Artificial Analysis tested the model at max effort. M3.1 Flash scored 42 on its Intelligence Index, 13 points above M3, level with GLM-5.3-Flash and behind DeepSeek V4.1 Flash 1017 at 45. It used about 189 million output tokens to run the index, fewer than those two models, which put its cost at about $0.22 per index task against about $0.51 for M3 and $0.26 for DeepSeek V4.1 Flash 1017. On AA-Omniscience it answered 56% of questions correctly, but because it attempted almost every question its hallucination rate was 93%.

Arena placed the model 12th among open models and 32nd overall in Code Arena WebDev with 1,588 points, 106 points above M3.

Reuters reported the launch from MiniMax's statement. MarkTechPost noted that every benchmark figure is vendor-reported and that MiniMax has not explained how it arrives at "1/3 the cost of M3", since the per-token price is half.

What We Don't Know Yet

  • The licence for the M3.1 Flash weights, and whether they ship on October 23 as the X post says.
  • Parameter count and architecture. MiniMax has not published a technical report or model card for M3.1 Flash.
  • How MiniMax calculates the 70% lower task cost and the "1/3 the cost" figure.
  • Independent results beyond Artificial Analysis and Arena.
  • When the Pro model will launch.

Sources