MiniMax M3.1 Flash in practice: tests and comparisons with DeepSeek V4.1 Flash

MiniMax confirms Space Bunny was an early M3.1 Flash, publishing scores (Automation-Bench 62.2%, DeepSWE 73.7%), $0.15/$0.60 API pricing, 1M context and Oct 23 open weights.

Same anime-cabin prompt: M3.1 Flash still can't beat DeepSeek V4.1 Flash 1017, but shows clear improvement over M3.

The author notes the free Space Bunny model in OpenCode is confirmed as MiniMax M3.1 Flash and still costs $0, urging readers to point it at a real repo before the free window closes, with setup in the reply.

MiniMax confirms Space Bunny was an early M3.1 Flash that processed over 200 trillion tokens; pricing is $0.15/$0.60 per million tokens, about 1/30 of Opus 5.5, with open weights due Oct 23.

A MiniMax staffer says M3.1 Flash delivers top-model performance on agentic tasks, real-world coding and computer use benchmarks at a very low per-task cost.

A user says MiniMax M3.1 Flash, formerly the stealth model Space Bunny, beats DeepSeek V4.1 Flash 1017 on parts of a benchmark and will be open-sourced.

After paying for a 49-yuan plan, the user judges MiniMax M3.1 Flash to be on the same level as GLM-5.3 Flash and offers to run requested test cases instead.

A summary of the official launch: M3.1 Flash targets agent, coding, computer-use and security work, with nearly 3x faster inference, over 200 TPS, DeepSWE v1.1 at 73.7% and OSWorld V2 at 61.7.

The author confirms Space Bunny was an early MiniMax M3.1 Flash and is testing it in MiniMax Code, which starts at 512K context with a 1M toggle and effort levels from low to xhigh and max; the team says the final version codes better than Space Bunny.

The author says MiniMax confirmed M3.1 Flash's open weights will be released this week, just a few days away.

A rundown of official specs — 1M context, text/image/video input, five effort levels defaulting to max, thinking always on — plus hands-on notes that frontend work nears flagship level while complex interactions and state management remain shaky, with speed and value as clear wins.

The author says they enjoyed testing Space Bunny, is curious how it scores on Artificial Analysis, and asks others about their experience.