The model is available through the MiniMax API, M Plan and MiniMax Code. The model weights have not yet been released: MiniMax said they will arrive October 23.
Faster inference at lower posted prices
MiniMax describes M3.1 Flash as a “generational leap” in agentic tasks, coding, computer use and cybersecurity. The company says online inference is nearly three times faster than M3 and that agentic-task usage costs are 70% lower.
Its posted API rates are $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached-input tokens. MiniMax also says output can exceed 200 tokens per second, with median speeds above 170 tokens per second.
The company reports scores of 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents’ Last Exam and 82% on CyberGym. Those are vendor-reported results, not independent reproductions, and the benchmarks cover different kinds of coding, computer-use and security tasks.
The model previously went by Space Bunny
M3.1 Flash is the public identity of Space Bunny, a model previously offered anonymously through OpenRouter. OpenRouter said Space Bunny ranked first on its service for 10 straight days and processed 56.7 trillion tokens before leaving October 5 and returning under the MiniMax name.
A separate external signal came from Arena’s Code Arena WebDev ranking, where M3.1 Flash debuted at No. 12 among open models and No. 32 overall with 1,588 points. That result covers Arena’s web-development test rather than the model’s full range of claimed capabilities.
