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

The poster confirms Space Bunny was an early M3.1 Flash build, cites official scores of 74% DeepSWE, 91% BabyVision and 62% Automation-Bench, and says hands-on results were solid at one-third M3's cost.

A side-by-side comparison finds M3.1 Flash behind DeepSeek V4.1 Flash on AA index (42 vs 45) and Terminal-Bench (38.1 vs 42.3), but cheaper per task ($0.22 vs $0.26), faster (68 vs 22 tok/s), flat-priced and able to take video input.

Both sit at $0.15/$0.60 and are strong on coding and agents; MiniMax claims higher speed and Automation-Bench, while DeepSeek has cheaper cache hits and edges several DeepSWE/Terminal benches.

A hands-on note says M3.1 Flash's reasoning_effort only ranges low to max with max as default, so cheap agent loops should set it low; $0.15/M input, 1M context, free on OpenCode for 3 days.

A tester reports M3.1 Flash scoring 42 on the intelligence index and asks how fast the model actually runs.

Hands-on with Space Bunny/M3.1 Flash: strong paper scores (74% DeepSWE) but weaker than GLM 5.3 Flash for everyday coding in OpenCode, verbose and context-hungry, more implementation mistakes than expected; paid users report better results than free-endpoint users.

The author says they used the anonymous Space Bunny for a while, reporting 74% on DeepSWE, 91% on BabyVision, 3x faster than M3 at a third of the cost.

The author had stealth Space Bunny rebuild a full landing page from a 40-second screen recording, now confirmed as early M3.1 Flash, which MiniMax says scores 74% on DeepSWE and 91% on BabyVision at 3x speed and a third of M3's cost.

Hands-on test: running the pelican prompt at High reasoning during OpenCode's 3-day free window, the author finds detail solid but overall weaker than deepseek-v4.1-flash-1017.

The author says he tested Space Bunny anonymously on OpenCode weeks ago: fast, strong results, above Claude Opus 5 on the leaderboard, 74% DeepSWE and 91% BabyVision, at 1/3 M3's cost and 3x speed, with weights due October 23.

A roundup of official M3.1 Flash details (1M context, $0.15 in / $0.60 out, $0.015 cache reads, DeepSWE 73.7) noting a third-party comparison where it trails DeepSeek V4.1 Flash on 4 of 6 shared benchmarks.

OpenCode confirms Space Bunny was M3.1 Flash: it led the board 12 straight days, served 106T tokens by Oct 5, about 1.7x DeepSeek V4.1 Flash, and is free for three days.