Le Chaton Fat is a model too fat to be true

If you spent any time on Reddit or X this weekend, you ran into Le Chaton Fat. Billed as Mistral AI's new cutting-edge model, the thing crushed Claude Mythos on the first tests, posted scores nobody had ever seen, and promised all by itself to put Europe back on the AI map¹. One small detail spoiled the party. It never existed.

A kitten born from a renamed cat

It starts with a funeral. At the first summit Mistral held in Paris last month, CEO Arthur Mensch announced that its assistant, Le Chat, would be renamed Vibe². On Reddit, some of the faithful went into mourning for what they considered one of the best product names in the business. Then, grief being what it is, they started inventing feline successors, each more ridiculous than the last. That's where Le Chaton Fat came from, a name that mashes French and English purely for the joy of contradicting itself, since a chaton is a kitten and fat is, well, fat¹.

It escaped Reddit on a Saturday and swept across X within hours, becoming one of the most talked-about topics in France and beyond. Mistral's CEO even joined in, with a deadpan "It's actually le gros chaton"¹. Typo fix or blessing? Your call. The meme leaked well past the AI crowd, to the point where Ethan Mollick, who co-directs an AI lab at Wharton, said he expected to be asked in meetings about "Mistral's new ginormous cat model with infinite benchmark scores"².

Benchmarks worth a laugh

What makes the hoax great is how straight it plays it. The fake spec sheets use the exact vocabulary of real launches. You read about a thirty-trillion-parameter model³ with room for a million tokens of context, able to handle images as well as text. And on top of that, a 90.6 on the MMLU-Pro test, while the very real Claude Fable 5 tops out at 88.9¹. Tucked between the believable numbers are the giveaways, a throughput of "1,000 meows per second" and a glorious "maximum chonk" listed as a feature².

Two fake websites keep the bit going with impressive composure⁴. The French one boasts twenty-four trillion parameters, a frank throughput of "0 tok/s," and access "reserved for French citizens," before admitting in the footer that it's a "satirical site." The English one keeps a flawless product-page voice, all the better to pretend it beats Claude Opus 4.8 and GPT-5.5, comparison table included. I laughed.

Under the laughs, there's a real critique, and the AI crowd happily adopted it. Benchmarks⁵ often don't mean much. A model can top a leaderboard nobody's heard of, on tasks nobody actually uses, and the announcement still gets treated like a geopolitical event. Le Chaton Fat just pushes that theater all the way to absurd. The hunger behind it is real, though, because everyone is waiting for a model that's open, affordable, and strong enough to actually matter².

When the joke turns on Europe

The joke gets less innocent when it aims at Europe. Fake press releases claimed the European Union had suspended Le Chaton Fat on security grounds, finding it "too heavy to regulate." Others swore Mistral had pushed it live by mistake and yanked it in a panic¹. You'll recognize the reference. It's almost word for word what happened to Claude, switched off by Washington a few days earlier. While America locks down its most powerful models, Europe dreams up a kitten so fearsome it has to be quarantined.

The jab was also aimed at the politicians. Emmanuel Macron's line about "the only model able to rival the American and Chinese giants" got passed around a lot, repurposed by people who see Le Chaton Fat as the dream model of a Europe still well short of the mark¹. Easy mockery, sometimes in bad faith, but it hits where it hurts.

The laugh and the lag

It's not all doom and gloom. Mistral doesn't need to beat Claude Mythos to exist. The company has chosen to polish smaller models like Mistral 3 and Ministral and to sign contract after contract with European businesses, where its resources actually go far¹. Its silence over the past few weeks is feeding the rumor mill, and a real announcement may be on the way, no cat involved.

What gets me isn't the joke so much as what it reveals. If Le Chaton Fat got this many laughs, it's because it scratches a very real itch, the wish for a European champion that can play with the big kids. A wish so strong that part of the internet almost believed in a cat. You can smile at that, as long as you remember the lag is no joke. Europe would do well to fund its models with the same energy it spends making memes about their absence. Meanwhile, I'll admit to a guilty soft spot for this imaginary fat tabby.

Notes

1. Numerama, Nicolas Lellouche, "C'est quoi Le Chaton Fat, le « modèle » de Mistral AI qui affole les réseaux sociaux ?" (in French), June 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

2. Business Insider, Thibault Spirlet, "The 'Le Chaton Fat' meme techies can't stop talking about", June 2026. ↩ ↩² ↩³ ↩⁴

3. Parameters are a model's internal settings, and their count gives a rough sense of its size. Tokens are the chunks of words it reads and writes. The fake sheet even specifies "256 experts," specialized sub-models of which only a few work on any given word. Real models do that too, which is why it all sounded plausible. ↩

4. The parody sites lechatonfat.com (in French) and lechatonfat.net, accessed June 2026. ↩

5. A benchmark is a battery of standardized tests used to compare models. MMLU-Pro, for instance, lines up thousands of multiple-choice questions across dozens of subjects. Think of it as the SAT, with the same problem, which is that if you cram long enough, you can ace the exam without being able to do much. ↩