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

A post pitting MiniMax M3.1 Flash against DeepSeek V4.1 Flash, with the body just a title and a link to the original comparison.

The author tried the free MiniMax M3.1 Flash with Claude on a pelican-running-a-marathon animation and found the result needed prompt tuning.

After a few hours of testing, the model is a big step up from M3 with strong computer use and Three.js results that sometimes beat DeepSeek first try, but it hallucinates and goes off-script at xhigh/max and feels slow despite 100+ tok/s.

A user testing M3.1 Flash via OpenCode found it stuck in a web search loop, talking a lot and repeating itself, though very thorough.

The author revisits their own one-shot tests with Space Bunny, still impressed and having expected a Pro-tier model; they expect the official release to be stronger, call it good value under MiniMax's token plan, and look forward to M3 Pro.

Hands-on test: a prompt asking M3.1-Flash and MiniMax Code to make a self-introduction video using built-in image and music generation took 41 minutes 27 seconds.

With M3.1 Flash weights due Oct 23, the poster's first impressions lead him to decide against upgrading his local Spark cluster to run it.

MiniMax officially priced M3.1 Flash at $0.15 input / $0.60 output per 1M tokens, reported 73.7% on DeepSWE v1.1 and 61.7 on OSWorld V2, confirmed Space Bunny was an early build, and said weights come later with no date.

Relays the official launch: Space Bunny was MiniMax M3.1 Flash, claimed nearly 3x faster online and about 70% cheaper for agentic work, priced at $0.15/$0.60 per million tokens and live via the MiniMax API, M Plan and MiniMax Code, with open weights to follow.

At max reasoning on the same prompt, MiniMax M3.1 Flash took 40 minutes at $0 for a bright daytime city, while DeepSeek V4.1 Flash 1017 took 50 minutes at $0.4 for a dusk fantasy look; a clear gain over the older M3.

The author lists M3.1 Flash at $0.15/$0.60 per million tokens and compares it with DeepSeek V4.1 Flash, Kimi K3 and GLM-5.3, arguing the flagship price war has begun.

Relays MiniMax's official figures: 73.7% on DeepSWE, 170+ tok/s median output, $0.15 input / $0.60 output per 1M tokens, claimed 3x M3's speed at a third of the cost; the author says they want to test it themselves.