Mistral Large 4 Brings 1.05T Parameters to European Infrastructure

Europe just put a trillion-parameter model on the board. Mistral's newest release, Mistral Large 4, carries the nickname Le Chonk and runs 1.05 trillion total parameters with 49 billion active at once. The API is live now, and the open weights are expected by the end of October.

That is a frontier-class model you can call today, with a downloadable version arriving within weeks. The pitch is not just size — Mistral says it built and serves the whole thing on European infrastructure.

What Happened

• 1.05 trillion total parameters, with 49 billion active per token.
• Native multimodal input.
• Built and served on European infrastructure.
• API available now. Open weights expected by the end of October.

Mistral calls it the strongest open-weights model out of Europe on aggregated benchmarks.

It lands in a crowded field. DeepSeek is pushing trillion-parameter-scale architectures, Kimi is playing in the multi-trillion-parameter range, and GLM is going after coding and agentic workflows. Different labs, different bets, same direction.

Why It Matters

A trillion parameters does not mean a trillion switch on for every token. These are sparse Mixture-of-Experts models, so only a fraction of the parameters activate per token — which is how something this large stays practical to run at all.

API access is not the same as open weights, and open weights are not the same as easy self-hosting. Once the weights land, people can fine-tune them, run them locally, or build custom agents on top. The real test is performance, inference cost, latency, and how much hardware it takes. Bigger numbers alone do not win the race.