Paris-based Mistral released Mistral Large 4, nicknamed "Le Chonk," its largest model featuring about one trillion parameters with 52 billion active per step. The model handles text and image inputs while outputting text and code under a modified MIT license. Available on Hugging Face, it requires substantial hardware such as eight Nvidia B300 GPUs for the standard version or four GB300 GPUs for a compressed variant.
Following additional reinforcement learning, Large 4 achieves strong coding performance, scoring 41 percent on Terminal-Bench 4.0 and 48 percent on Artificial Analysis' Coding Agent Index. It runs at 161 tokens per second with lower API prices and records a 96.6 percent refusal rate for harmful cyber requests. However, it trails top models from Anthropic and Google as well as several open Chinese models like DeepSeek V4.1 Flash and GLM-5.3 in overall benchmark rankings.
This release matters because it provides enterprise users with a fast, capable open-weights coding model that runs on local hardware, while highlighting the fierce global competition between European and Chinese AI developers.