MiniMax releases M3.1 Flash model

On October 19, 2026, MiniMax announced M3.1 Flash on its official X account, describing it as reaching frontier-level intelligence in agentic AI scenarios and as redefining the performance-to-cost ratio for that field. The company also said Space Bunny, which had topped global leaderboards after processing more than 200T tokens anonymously, was an earlier version of M3.1 Flash. A post on Reddit's LocalLLaMA community said the model had been released and that its weights would be made available on October 23, without offering further details on parameters, capabilities or release format.

Coverage from 36Kr and IT之家 the same day added specifics: M3.1 Flash is said to bring generational gains in agentic tasks, coding, Computer Use and cybersecurity, with online inference roughly 3x faster and usage costs 70% lower. MiniMax also teased a Pro version aimed at a higher intelligence ceiling, to arrive soon. IT之家 reported that the model reaches leading-model levels in complex agent tasks (Cowork), Computer Use, software development and cybersecurity, ranks at the front of AutomationBench-AA, and scores 76.2 on BankerToolBench.

A V2EX user post offered a different set of figures: 73.7% on DeepSWE and 61.7 on OSWorld V2, output speeds of 200+ TPS, and pricing of 1 yuan per million input tokens, 4 yuan per million output tokens and 0.1 yuan for cache, with weights to be open-sourced later. The same post framed the release as aimed at ds4.1flash-1017 and predicted more models in the coming days, which are the poster's own assessments. Public information currently stops at launch: the only stated date for weights is October 23, and no specific date has been given for the Pro version.