MiniMax M3.1 Flash was officially released on October 19, 2026, in a MiniMax blog post. The new-generation Flash model targets agentic tasks, coding, computer use and cybersecurity: online inference runs nearly three times faster than its predecessor, while overall usage cost for agentic tasks drops by 70%. MiniMax also confirmed that Space Bunny, the anonymous early version that processed more than 200 trillion tokens, was M3.1 Flash.
The scorecard focuses on tasks that do real work
In MiniMax's published evaluations, M3.1 Flash ranks among the top performers on AutomationBench-AA for enterprise automation. On BankerToolBench it scores 76.2, testing multi-step planning, consistency across documents and files, financial compliance and chained tool calls. For coding, it reaches 73.7% on DeepSWE v1.1, where models must independently understand code, build features, fix issues and validate through testing. For computer use, it scores 61.7 on OSWorld V2 and 29.6 on Agents' Last Exam, which MiniMax says are both first place among Chinese models. For cybersecurity, it scores 82% on CyberGym, and internal security experts also evaluated code auditing, vulnerability patching, penetration testing, malware analysis and reverse engineering. On speed, MiniMax says everyday output can reach 200+ TPS, with median speeds consistently at 170+ TPS, thanks to more aggressive sparsity.
The price list states the positioning plainly
M3.1 Flash is priced at $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. It is available now through the MiniMax API, M Plan and MiniMax Code; model weights will be released later, and a Pro version pushing the intelligence frontier will also arrive later. This price band, paired with near-top scores, is aimed not at casual chat users but at developers who run agents for long stretches and process tasks in bulk: for such workloads, the bill is dominated not by the model with the highest unit price, but by the total cost of completing each task.
The Flash race is now about cost per completed task
One caveat: all of the scores above are vendor-published results that have not yet been independently reproduced at scale, so a small trial on real tasks is still worthwhile before wiring it into a workflow. But the direction is clear. When Flash-tier models can reach the top tier on coding, computer use and security tasks, competition shifts from parameters and leaderboard places to who can actually finish tasks at the lowest total cost. By pricing M3.1 Flash at this level, MiniMax is forcing rivals to answer the same question.
