In late May, Mistral renamed its chatbot Le Chat — French for “the cat” — to “Vibe.” The fans mourning the cat did what the grieving do online: they built a replacement. Le Chaton Fat. Franglais for “the fat kitten,” an oxymoron with a face. It started in Mistral’s own community, hit X around June 11, and within hours actual machine-learning researchers were asking out loud if it was real. Cody Blakeney posted, dead serious, “Can someone tell me if Le Chaton fat is real or an amazingly elaborate joke?” No weights. No API. No model card. It never hit a real leaderboard. The only chart anyone saw was one somebody drew as a bit.
And the specs kept getting fatter. It launched at 24 trillion parameters — a number Mistral itself floated in a since-deleted post — then the crowd shoved it to 30 trillion, then 100 trillion on the comic charts. Throughput: “1,000 meows per second.” Context window: infinite, “measured in croissants.” It beat a made-up benchmark called VoltaireBench by 30 points over Anthropic’s Fable 5 and, for safety, would only write code in French. All of it absurd, all of it close enough to a real launch that nobody could call it.
That’s the whole joke, and it was never on the people who fell for it. It was on an industry where “100 trillion parameters and infinite context” reads like a Tuesday. Mistral knew it — CEO Arthur Mensch “corrected” everyone that it was actually “le gros chaton,” and the company quietly slapped a chubby cartoon cat on Vibe. When you can’t tell the parody from the product, the parody won.