Paris-based Mistral releases "Le Chonk," a strong coding model that still trails China overall
decoder ·
Paris-based Mistral has released the weights for its largest model, Large 4, nicknamed "Le Chonk," which features approximately one trillion parameters and operates under a modified MIT license. While the model demonstrates strong coding capabilities and high speed at 161 tokens per second, independent benchmarks indicate it trails behind leading Chinese open-source models like DeepSeek V4.1 Flash and GLM-5.3 in overall performance. On Artificial Analysis' Intelligence Index, Large 4 scored 26 points, placing it significantly behind top proprietary systems from Anthropic and Google as well as superior Chinese competitors. The model excels in cost-efficiency compared to some rivals and boasts a high refusal rate for harmful cyber requests, though it ranks third among open models in cyber defense tasks. Mistral emphasizes its value as a sovereign AI solution trained entirely in Europe, allowing for local deployment on hardware such as Nvidia B300 GPUs. Despite these strengths, the company acknowledges that its resource constraints prevent it from reaching the top tier occupied by US and Chinese labs with greater capital. The release marks a strategic move to provide a robust, affordable alternative for enterprises seeking control over their AI infrastructure.
decoder