Open weightsMistral
Mistral published the weights of Large 4, a one trillion parameter model, under a modified MIT licence
47% fewertokens per task than GLM-5.3 at the same 48% on the Coding Agent Index, Mistral's claim
Mistral released the weights of Mistral Large 4, nicknamed Le Chonk, on 29 October, three weeks after a preview. It is a multimodal mixture of experts model, one that uses only part of its parameters per token: 1 trillion in total, 49 billion active, with a 1M token context. Mistral says it scores 48% on the Artificial Analysis Coding Agent Index, tied with GLM-5.3, and 41% on Terminal-Bench 4.0. Anyone selling the model as a service needs a separate licence. Benchmarks are Mistral's.