What changed
MiniMax announced M3.1 Flash in a blog post dated October 19, 2026. It is a Flash-tier model aimed at agentic tasks, coding, computer use and cybersecurity. MiniMax says an early version ran anonymously as Space Bunny and processed more than 200 trillion tokens.
The company says it made inference nearly three times faster through more aggressive sparsity, with output speeds above 200 tokens per second and a median above 170 tokens per second, and that overall usage cost for agentic tasks is 70% lower than its predecessor.
Who can use it and when
M3.1 Flash is available now through the MiniMax API, the M Plan subscription and the MiniMax Code desktop app. MiniMax's API documentation lists a 1M-token context window. OpenRouter offers it as minimax/minimax-m3.1-flash at MiniMax's list price; a free launch variant ran until October 22, 2026. MiniMax says the model weights will be released later and that a stronger Pro model will also follow, without giving dates.
Why it matters
MiniMax reports 73.7% on DeepSWE v1.1, 61.7 on OSWorld 2.0, 29.6 on Agents' Last Exam, 76.2 on BankerToolBench and 82.2% on CyberGym. These are the company's own results, and its own table shows that M3.1 Flash does not lead everywhere: DeepSeek V4.1 Flash-1017 scores 77.0 on DeepSWE and Opus 5 scores 74.0.
The main argument is cost. On MiniMax's Automation-Bench chart, M3.1 Flash reaches 52.8% to 62.8% depending on effort, at an average cost per task below $0.05, while Opus 5.5 and GPT-5.6 cost several times more per task in the same chart. If your agents run many long tasks, that gap may matter more than a few benchmark points.
Access and limitations
API pricing is flat: $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. MiniMax ran most of its evaluations with its chosen harnesses, such as Claude Code and mini-SWE-agent, so your results with a different agent setup may differ.
The weights and their licence have not been published yet, so self-hosting is not possible today. AppFor.ai has not tested the model. Run it on a sample of your own tasks and compare quality and cost with the model you use now before switching.
Go to the source
Sources & further reading
Based on the official documentation below. Features and access can change; check the provider before making a decision.
Was this page useful?
Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.