Paris-based Mistral AI launched Mistral Large 4 on Tuesday, introducing a 1 trillion-parameter model that the company is positioning as a top-tier open-weight system. Parameters are the adjustable values an AI model tunes during training, and larger counts usually mean more learning capacity as well as higher operating costs.
On X, Mistral AI chief scientist Guillaume Lample wrote that Mistral Large 4 is “at the frontier of open models, and by far the strongest open-weight model from the US or Europe.”
Mixture-of-experts design keeps active parameters lower
Not all 1 trillion parameters are used at once. Large 4 uses a mixture-of-experts architecture, routing each prompt to a smaller set of specialist subnetworks. As a result, only 49 billion parameters are active for each answer.
Mistral’s previous flagship, Large 3, used the same approach. That model had 675 billion total parameters, with 41 billion active per response.
How the “Le Chonk” name emerged
The “Le Chonk” label comes from a community joke that took off in June after Mistral renamed its Le Chat assistant to Vibe. Users on Reddit and X invented a fictional model called Le Chaton Fat, roughly meaning “the fat kitten,” and gave it absurd specs: 30 trillion parameters, 1,000 meows per second, and fake benchmark claims saying it beat Claude Fable 5.
CEO Arthur Mensch joined in, replying that the model was actually called “le gros chaton,” French for “the big kitten.” Mistral later added a cartoon cat to the Vibe website. In the real launch post for Large 4, the company now lists the model as, “very officially,” le Chonk.
Sovereign AI pitch and recent funding
Mistral sells what it calls “sovereign AI,” meaning models that a country or company can own and run itself without handing data to outside firms. According to the report, Saudi Arabia’s state-backed HUMAIN signed a deal worth hundreds of millions of euros with Mistral in August for that reason.
The company also raised a €3 billion ($3.37 billion) Series D in September at a valuation above €21 billion ($23.6 billion), with Samsung leading the round. Mistral said Large 4 is the first milestone on the roadmap funded by that capital.
Pricing undercuts Claude Opus 5.5 and GPT-6 Astra
Mistral listed Large 4 at $1.36 per million input tokens and $4.18 per million output tokens. Tokens are the units AI companies bill for, and the report describes them as chunks of text roughly equal to three-quarters of a word.
Claude Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens. GPT-6 Astra is priced at $10 and $50, respectively.
That puts Large 4 at about one-third of Opus 5.5’s input price and about one-fifth of its output price. Against Astra, the model comes in at roughly one-seventh on input and one-twelfth on output.
Benchmark comparisons focus on Chinese open-weight rivals
Mistral’s launch materials compare Large 4 mostly with Chinese open-weight models, including DeepSeek V4 Pro, Kimi K3, GLM-5.3, and Qwen3.8 Max. In this context, open-weight means the model can be downloaded and run by anyone.
Claude and GPT appear in only a small number of comparisons. The newest Claude model, Opus 5.5, shows up only in a cybersecurity-related claim.
Surge AI blind coding evaluation
In a blind human evaluation of coding quality conducted by Surge AI, Mistral Large 4 ranked second out of five models with a score of 3.74 out of 5. Claude Opus 5 scored 4.22 and ranked first.
AutomationBench results
AutomationBench gives an AI system 657 tasks inside simulated business software covering finance, HR, sales, and support. It scores the share of each task’s goals completed, and gives zero credit if the model breaks a rule. Large 4 scored 59.9 on that benchmark.
On Artificial Analysis’s board for the same test, Claude Sonnet 5.5 scored 71.8, Opus 5.5 scored 69.5, and Gemini 4 Argon led with 77.5.
DeepSWE 1.1 coding benchmark
On DeepSWE 1.1, a benchmark developers use to assess coding ability, Large 4 scored 62. That was higher than GLM-5.3 at 61 and DeepSeek V4 Pro at 57, but below Kimi K3 at 68.
Datacurve’s own leaderboard lists GPT-6 Astra and Claude Opus 5 at 74.
Weights planned for release by late October
Mistral said it will release Large 4’s weights by the end of October. That would allow outside developers to download the model and test the company’s performance claims on their own.