Paris-based Mistral releases "Le Chonk," a strong coding model that still trails China overall

Mistral has released the weights for Mistral Large 4 on Hugging Face, three weeks after its public preview. The company nicknamed the model "Le Chonk." It accepts text and images as input and outputs text and code. The model has about one trillion parameters in total, with 52 billion active per step. According to the SGLang docs, the standard version runs on eight Nvidia B300 GPUs, and a more heavily compressed variant runs on four GB300 GPUs.

Mistral says it kept training the model with reinforcement learning after the preview, with large gains in coding and cyber defense. Researchers Pierre Stock and Corentin Petit described the process on a press call. The company said performance is limited only by how hard a task the team can still create and verify.

On Terminal-Bench 4.0, Large 4's score rose from 28 to 41 percent, just ahead of GLM-5.3 at 40 percent. Mistral says the model reaches 48 percent on Artificial Analysis' Coding Agent Index, nearly tied with GLM-5.3, but uses 47 percent fewer tokens. Claude Code with Claude Opus 5.5 leads that index at 62 points. On Cognition's FrontierCode 2.