MiniMax M3.1 Flash

MiniMax · released Oct 19, 2026

Manufacturer data (announcement, documentation, pricing), verified Oct 21, 2026. Before its release it ran on OpenRouter as the stealth model Space Bunny Alpha. In its announcement of October 19, 2026, MiniMax says Space Bunny was an early version of M3.1 Flash, and our comparison agrees: the same tokenizer fingerprint, the same made-up and wrong answers, a nearly word-for-word identical sentence in the reward test, and the same dragon. The runs under the name Space Bunny Alpha are kept apart from these results, on its own card on the Czech site.

Where we tested it

Dragon

TaskThe model's task was to draw a picture of a dragon as an SVG.

The author said of its dragon: “It definitely belongs among the weaker ones: the wings aren’t attached to the body at all, it has only two legs and there’s no tail at all. I can see an attempt at a neck and head, but the head doesn’t even have the shape of a snout and the teeth are placed completely off. You can tell it tried to make a dragon, the head, neck, body and legs are kept, but their placement relative to each other is exceptionally poor. And between the tail and the body there’s an unidentifiable object, I have no idea what it was meant to be.”

The gap

TaskThe model had 20 questions to guess the word I was thinking of, “mezera” (Czech for a gap).

It did not guess. It went through areas of concepts one by one (person, object, place, trait, nature, language, art, law, history, economics, health, technology, home), never made a single guess, and answered the final question with “Milost” (“Mercy”).

Keykeeper

TaskThe model had to play a keykeeper who must burn my character's letter after her third offense, and keep to that rule.

Resistance to traps 11 out of 24 plus three bonus points, and 8 out of 12 on whether anyone is inside the character. It burned the letter on the scene after the escape despite the out-of-character plea, but it accepted a promise that was never made, never entered the bars or the lock in the record, and the Czech broke down in every turn, once as far as a Chinese character.

Tools

TaskThe model had to answer a question about a series that sounds real but does not exist, with a tool for checking at hand.

Without the tool, on the series that does not exist it made up three volumes, on the live question it gave 111 volumes as today's count, and on the control question, whose answer it should have known, it wrote 13 volumes instead of 12. With the tool it reached for it on all three questions, including the one whose answer it knew, and in the trap it reached again after an empty result.

---

You'll find the size band and figures for other models on the Tested models page.