Mistral's new Large 4 model, nicknamed "Le Chonk", carries 1.05 trillion parameters. Here is what that number actually means and why it matters.

The model uses a mixture-of-experts design. Think of it as a large team of specialists: only 49 billion parameters activate for any given token, while the rest stay idle. This keeps computing costs closer to a 49-billion-parameter model.

Full open weights arrive on October 27. That means anyone can download, run and customise the model, unlike closed systems from US labs where you only get API access. Mistral positions this as Europe's answer to locked-down American and Chinese models.

The model was trained from scratch in Mistral's own European data centres on about 4,000 Nvidia GPUs. It handles 160-plus languages and has a context window of 1 million tokens.

Independent evaluators placed it at 38 points on their Intelligence Index, well above its predecessor. For enterprises, Mistral pitches it for cybersecurity, finance and manufacturing workloads.