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

The author ran a Japanese small-town train test with MiniMax M3.1 Flash; output was decent but the night train lights drifted away, and DeepSeek V4.1 Flash 1017 still did better.

Testing M3.1 Flash automation in Blender to model Tokyo Tower: decent output but below Opus 5.5, taking about 44 minutes 36 seconds and costing roughly $0.57.

A roundup noting MiniMax M3.1 Flash is free to use until the 22nd, that it was the anonymous Space Bunny with 1M context, and listing four free access channels.

On the same 3D Rubik's Cube task, DeepSeek V4.1 Flash finished in one generation for $0.26, while MiniMax M3.1 Flash needed five generations and $0.80.

The post breaks down the launch playbook: Space Bunny ran anonymously on OpenRouter on 200 trillion tokens of real traffic before the reveal, with agent costs down 70%, ~3x speed, and weights due around Oct 23.

The author recalls complaints during Space Bunny's anonymous run—overthinking, fast context burn, forgotten requirements on long tasks, issues found but not fixed—and asks how much the official release improved, while noting $0.15/$0.60 per million token pricing.

Lists M3.1 Flash pricing ($0.15 input, $0.60 output, $0.015 cached per million tokens) and near-3x M3 speed, and cites Artificial Analysis: it answers nearly every question, guessing on about 93% of ones it can't solve instead of admitting ignorance, with roughly 56% correct.

A video post says MiniMax M3.1 Flash has launched with local weights coming October 23, discussed alongside DeepSeek V4.1 Flash and other models.

The author says M3.1 Flash overtaking DeepSeek V4.1 Flash on leaderboards is being hotly debated, but urges patience, noting users benefit as capability rises without price hikes.

Space Bunny, tested anonymously for nearly a month on over 200 trillion tokens, is confirmed as an early MiniMax M3.1 Flash; API is $0.15/$0.60 with weights planned for Oct 23.

A developer notes MiniMax M3.1 and DeepSeek V4.1 cost about the same and both are strong at coding and UI, and asks which API people actually use.

In a price-performance index combining cost and capability, DeepSeek V4.1 Flash 1017 ranks first, followed by MiniMax M3.1 Flash and Claude Haiku 5.5.