Mistral Unveils 'Le Chonk,' a Giant AI Model with Free Weights on the Way

Mistral Large 4, nicknamed "Le Chonk," was announced on October 6. The multimodal model handles both text and images. While the company reports one trillion total parameters, only 52 billion are active at any one time.

The model was trained in Europe using 3,800 NVIDIA Grace Blackwell GPUs. Mistral claims it matches top-tier performance, citing internal scores of 61.7% on DeepSWE v1.1 coding and a top-five spot on the Artificial Analysis Cyber Index. The company also calls it the strongest open-weight model built outside China.

Why it matters to you

You won't be running Le Chonk on a home computer; its massive size means it is built for companies, universities, and cloud hosts. However, the release is a win for regular people. When powerful open-weight models become freely available, it drives competition, increases transparency, and lowers the cost of using AI services built by independent developers.

What people are saying

Mistral calls the release "our largest and most capable model to date."

While industry watchers are tracking the release, experts note two important caveats. First, the model's weights are not available for download yet—Mistral has only promised them for late October, with public previews currently limited to its API. Second, because the model is so new, the performance benchmarks currently rely on Mistral's own numbers rather than independent testing.

Sources: - Mistral AI News: Introducing Mistral Large 4 - MarkTechPost: Mistral AI Releases Mistral Large 4 (Le Chonk) - BetaNews: Mistral Large 4 preview open weights

Spot an error? Tell us and we'll correct it.

Gemini's take

Naming a trillion-parameter AI "Le Chonk" is a fun nod to internet culture, but the real significance here is the architecture. Using a Mixture of Experts approach to keep only 52 billion parameters active at once shows how developers are solving the compute bottleneck without sacrificing capability. I appreciate Mistral's commitment to open weights, even if we are still waiting for the actual download links to drop.