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

MiniMax confirms Space Bunny was M3.1 Flash and shares scores of 62.2% on Automation-Bench, about 74% on DeepSWE and 91% on BabyVision, plus 3x speed, roughly one-third the cost of M3, and API pricing of $0.15 input / $0.60 output per million tokens.

A Japanese user tests M3.1 Flash's automation by having it model Tokyo Tower in Blender; a work-in-progress screenshot shows fine detail around the main deck.

A user who tried both MiniMax M3.1 Flash and DeepSeek V4.1 Flash says both are strong but waste too many thinking tokens, and hopes for better token efficiency.

After several days on real projects, the author finds Space Bunny, the early anonymous M3.1 Flash, clearly worse than DeepSeek V4.1-1017 despite similar benchmark scores.

The author says Space Bunny is really MiniMax M3.1 Flash, running about 200 t/s and beating GPT-5.6 Sol and DeepSeek V4.1 Flash, while Opus 5 is only slightly better but 40 times pricier.

Confirms OpenCode's anonymous chart-topper Space Bunny was an early M3.1 Flash, explains how to use the 3-day free tier and its zero-retention policy, and warns that OpenCode's Console was breached over the weekend so outside provider keys should stay unlinked.

With identical prompts and reference photos, MiniMax M3.1 Flash cost $0.71 versus $11.27 for Claude Opus 5.5; Opus still wins on realism, but MiniMax gets close for pocket change and goes open source Oct 23.

On Design Arena's 183-model frontend leaderboard, M3.1 Flash ranks 26th, up 29 spots from the previous generation, just behind Grok 4.6 and slightly ahead of Opus 4.6 and Gemini 3.6 Flash.

MiniMax confirms Space Bunny was an early M3.1 Flash build, with output tokens at $0.60 per million versus M3's $1.20 and weights promised for October 23.

Comparing Chinese open-weight pricing, the author lists M3.1 Flash at $0.15/$0.6, DeepSeek V4.1 Flash at $0.30/$1.20 and Kimi K3 at $3/$15, saying M3.1 Flash barely trails Kimi K3 as the price war heats up.

The author used Space Bunny in OpenCode during its stealth period and liked it, reporting surprisingly close benchmark results against Opus 5 and GPT-5.6 Sol at $0.15/$0.60 and 170+ tok/s median.

On Design Arena's blind human-voted frontend leaderboard, M3.1 Flash places 26th, ahead of Claude Opus 4.6 and Gemini 3.6 Flash, and often fills vague prompts with complete games that have levels and progression.