Mistral AI Drops 'Le Chonk': A Massive AI Model Named After a Cat Meme

Paris-based Mistral AI on Oct. 6 unveiled Mistral Large 4, a 1-trillion-parameter AI model that uses a mixture-of-experts architecture and activates 49 billion parameters per query. The company nicknamed the release “le Chonk,” plans to publish the model weights by the end of October, and published benchmark results that place Large 4 ahead of GPT-6 Astra on one finance test but behind Claude variants on several other evaluations.

By AI Newsroom· Reviewed by Pranav, Founder & Editor-in-ChiefPublished about 1 hour agoUpdated about 1 hour ago0 views

Mistral AI Drops 'Le Chonk': A Massive AI Model Named After a Cat Meme

Why It Matters

The release marks a major open-weight model from a European AI firm at a time when sovereign and downloadable models are in demand, and Mistral’s decision to publish weights could let outside developers validate and build on its performance claims. Benchmarks and pricing also position Large 4 as a lower-cost alternative to some proprietary models, which matters for enterprises weighing on-prem or private deployments.

Key Facts

Launch date: Oct. 6, 2026
Model name: Mistral Large 4 (nicknamed “le Chonk”)
Parameters: 1 trillion total; 49 billion activated per query (mixture-of-experts routing)
Planned weights release: By the end of October 2026
Pricing (input): $1.36 per million input tokens

Mistral AI released Large 4 on Oct. 6, presenting a 1-trillion-parameter model that employs a mixture-of-experts design so only a subset of specialists are used per query—about 49 billion parameters active for each response. The company leaned into a community meme, adopting the nickname “le Chonk” that grew from a June joke around a fictional oversized kitten model; Mistral’s public materials now list the name playfully as “very officially, le Chonk.” Mistral described Large 4 as the firm’s latest flagship and said it will publish the model’s trained weights by the end of October, which would allow external developers and researchers to download and run the model themselves. The timing follows the company’s September €3 billion Series D at a valuation above €21 billion, a round Mistral said will fund its roadmap milestones including this release. In published comparisons, Mistral positioned Large 4 mostly against other open-weight models from China and Europe, while including a few head-to-heads with proprietary systems. On the Finance Agent v2 test, Large 4 scored 54.7, edging GPT-6 Astra’s 53.5 but trailing Claude Opus 5.5’s 58.6. Other benchmarks showed mixed results: Large 4 placed second in a human coding-quality evaluation (3.74/5 behind Claude Opus 5), scored 59.9 on AutomationBench against higher Claude and Gemini results, and posted a 62 on DeepSWE 1.1—above several Chinese open models but below some top proprietary entries. Mistral also highlighted pricing that undercuts several proprietary alternatives: Large 4’s listed rates are $1.36 per million input tokens and $4.18 per million output tokens, compared with examples Mistral provided such as Claude Opus 5.5 at $4/$20 and GPT-6 Astra at $10/$50. The company markets itself as a provider of “sovereign AI” that customers can run without sending data to external providers; it has previously secured large commercial deals citing that proposition. While Mistral’s results show strengths in particular tests and cost advantages on token pricing, the firm’s own benchmarks place Large 4 variably against different competitors depending on the workload. The forthcoming weights release will let independent researchers verify the model’s claims and evaluate its performance across a broader set of tasks.