Mistral Large 4 'Le Chonk' Is Here: a 1-Trillion-Parameter Open Model With Weights Landing by End of October

Mistral AI on Tuesday released Mistral Large 4, its new flagship model — and confirmed the internet's favorite joke as its official internal nickname, "Le Chonk." The model carries 1 trillion total parameters with a sparse mixture-of-experts design that activates 49 billion per query, and the French lab says downloadable weights will follow by the end of October after a preview period on its API.

The scale jump is substantial. Mistral's previous flagship, Large 3, ran 675 billion parameters with 41 billion active. Large 4 was trained from scratch — the company says without distillation from a competitor's outputs — on 4,000 Nvidia Grace Blackwell GPUs over roughly two months in Mistral's own European data centers, covering more than 160 languages including every official language of the European Union.

Pricing undercuts the closed frontier by a wide margin: $1.36 per million input tokens and $4.18 per million output tokens. For comparison, Claude Opus 5.5 charges $4 and $20 on the same metering, while GPT-6 Astra lists $10 and $50. That places Large 4 at roughly a third of Opus 5.5's input price and a fifth of its output price — about a seventh and a twelfth of Astra's.

The benchmark picture is mixed and, importantly, mostly vendor-reported and not independently verified. Mistral says Large 4 is the strongest open-weight model built in the US or Europe, with state-of-the-art open-model results on cyber defense, manufacturing and finance workloads, and that it beats closed frontier models on visual grounding. On DeepSWE 1.1 it scores 62, edging past Z.ai's GLM-5.3 at 61 and DeepSeek V4 Pro at 57 but trailing Moonshot's Kimi K3 at 68. A blind human evaluation of coding quality run by Surge AI placed it second of five models at 3.74 out of 5, behind Claude Opus 5's 4.22. Mistral's own launch chart shows it at 54.7 on Finance Agent v2, ahead of GPT-6 Astra's 53.5 but behind Claude Opus 5.5's 58.6.

Chief scientist Guillaume Lample framed the release as an argument about ownership rather than pure leaderboard position: "Sometimes, people like to make a big deal over the US, versus Europe, versus China. But what really matters is to own the model — even for US companies. If you use a closed model, there is no guarantee it will still be there tomorrow." The positioning leans on sovereignty: open weights let governments and companies run the model on their own infrastructure, including zero-data-retention configurations, instead of depending on a US provider's policies.

The nickname has a backstory. In June, after Mistral renamed its Le Chat assistant to Vibe, Reddit and X users invented a fictional model called "Le Chaton Fat" with absurd specs — 30 trillion parameters, 1,000 meows per second, fake benchmarks beating Claude. CEO Arthur Mensch played along at the time, and the company has now "very officially" adopted the oversized-cat reference for its real trillion-parameter model.

The launch is the first milestone funded by Mistral's September Series D — €3 billion (about $3.4 billion) at a valuation above €21 billion, led by Samsung, the largest private tech raise in European history and a round this site covered last month. Whether open weights can keep closing the gap on coding and reasoning will become testable the moment those weights land; Mistral has invited exactly that scrutiny.