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

The author says the official M3.1 Flash scores prompted an internal benchmark run via API, with DeepSeek 1017 still the best so far and results pending.

M3.1 Flash (Space Bunny) is free for 3 days on OpenCode at $0.15 input / $0.60 output / $0.015 cache read per 1M tokens, about half DeepSeek V4.1 Flash's price, with some tasks near GLM 5.3.

The author ran the same Bowser prompt on both Flash models and found MiniMax M3.1 Flash free and better, while DeepSeek V4.1 Flash cost $0.06 and came out worse.

Benchmark list for M3.1 Flash: Automation Bench 62.2% vs Opus 5's 50.3%, DeepSWE 74%, BabyVision 91%, BankerToolBench 76.2 tying Opus 5; open weights Oct 23.

A user test found both models took 38 minutes on the same task, but MiniMax's output was preferred, using under 1% of the weekly quota on a $22 plan.

An opencode user reports Space Bunny handles skills poorly — not reading or invoking them — and shows subtle hallucinations in long sessions with repeated compaction, needing very direct prompts.

A user ran the same prompt at max reasoning effort on both models: MiniMax M3.1 Flash took 40 minutes at $0, while DeepSeek V4.1 Flash-1017 took 50 minutes at $0.40, with a claimed big jump over the previous M3.

Confirms Space Bunny is MiniMax M3.1 Flash and lists DeepSWE v1.1 73.7, NL2Repo-Bench 61.2 and Terminal-Bench 4.0 38.1, all ahead of K3 and GLM-5.3 and near GPT-5.6 Sol.

A user testing MiniMax-M3.1-Flash in DeepSeekHarness reports hours of smooth runs with no hiccups or partial breaks, calling the cross-model compatibility impressive.

The author expected MiniMax to have pivoted to voice and video; M3.1 Flash feels a bit underwhelming after DeepSeek V4.1 Flash 1017, but rumored Pro-tier and 500B+ models are what they look forward to.

Roundup of M3.1 Flash pricing (¥1/¥4 per million tokens) and four free access routes; on the same pelican-on-bicycle prompt at max effort, DeepSeek took 611s for ¥0.63 and drew it right, while MiniMax took 1053s free but drew an egret.

Space Bunny is confirmed as an early MiniMax M3.1 Flash build; the post cites a 62.2 agent benchmark versus Opus 5's 50.3 and GPT-5.6 Sol's 45.8, plus $0.15/$0.60 pricing and Oct 23 weights.