Before you reshare the next screenshot of a “frontier model,” run it through the Le Chaton Fat test. Is there a real model card or product page, or just a screenshot of a chart? Can you actually call it — open weights to download, or an API key that works? Is the benchmark published and reproducible, or is the “leaderboard” one image somebody made in an afternoon? Le Chaton Fat failed all three: no weights, no API, no model card, and the only leaderboard it ever topped was a parody chart bragging about a croissant context window and 1,000 meows per second. Any model that can’t answer those three is a kitten.
The good news is the joke summoned something real. Somewhere between his “le gros chaton” wink and the cartoon cat mascot, Arthur Mensch teased an actual model — “fat indeed, but sparse,” the start of a new family, with early access opening in July for research, government, and industry partners. And it’s open-weight. Mistral’s own pitch: “you cannot own, inspect, audit, or improve a system you are only permitted to reach through someone else’s interface.” That’s the whole test in one sentence. Open weights mean you never have to take a chart on faith.
So where does June 25 leave us? Still waiting. The fat-but-sparse model has no name, no parameter count, no ship date past “this summer” — and the only fresh Mistral drop this week was OCR 4, a document reader with nothing to do with the teased family. Fine by me. The cure for hype is the cure for the meme: wait for the weights. When they land, download the thing and benchmark it yourself instead of arguing with a screenshot. Until then, every “frontier model” that flunks the test is a fat kitten with a good publicist.