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

A video demo of M3.1 Flash-generated frontend pages finds little visual difference from DeepSeek's output, with the poster praising domestic models' value.

On one HTML prompt for a submarine chase through an undersea city, MiniMax M3.1 Flash cost ~$0.21 in 29m24s, about 7x cheaper than Gemini 3.8 Flash and 1.8x cheaper than DeepSeek V4.1 Flash, while Claude Haiku 5.5 came in at ~$0.17 but slower.

On BugBunny's VulnPR-100 with one hour per PR, M3.1 Flash caught 35/100 for $11.85 ($0.34 per bug), matching Mistral Large 4 at ~4.5x lower review cost, while DeepSeek V4.1 Flash 1017 found 19 more for less.

MiniMax officially launched M3.1 Flash, claiming a generational leap in agentic tasks, coding, computer use and cybersecurity, with model weights due October 23.

A test gives MiniMax M3.1 Flash a front-end prompt to build an HTML page with an SVG 2D animation of a pelican riding a bicycle, with video of the result.

The anonymous model Space Bunny held #1 on OpenRouter for ten straight days and turned out to be MiniMax M3.1 Flash, priced at $0.15/$0.60 per million tokens with 72 hours free and open weights on Oct 23.

The author confirms Space Bunny was MiniMax: M3.1 Flash scores 62.2% on automation-bench ahead of Opus 5 and Sonnet 5.6, matches them on DeepSWE, runs 200+ tok/s, and costs $0.15/$0.60 vs Opus 5's $5/$25.

Recaps the official specs: 74 on DeepSWE, $0.15 input and $0.60 output per million tokens, one-third the cost and 3x the speed of M3, open weights on October 23, plus a rumored Pro model.

A user ran MiniMax M3.1 Flash for free on an Argon 4 comparison task; output quality was surprisingly good, but it took almost two hours to finish.

The author used M3.1 Flash with a Need for Speed: Most Wanted prompt to produce a neon racing demo, judging it maybe not the best game-building model but a solid result.

On the same candy-bear prompt, MiniMax M3.1 Flash took 57 minutes and produced usable but clearly weaker output than DeepSeek V4.1 Flash 1017.

Recap of anonymous model Space Bunny drawing 200T+ tokens from users who picked it blind, revealed as an early MiniMax M3.1 Flash; the release claims 74% on DeepSWE and 91% on BabyVision at $0.15 per 1M input tokens.