French AI lab Mistral has released Mistral Large 4, a 1-trillion-parameter multimodal model nicknamed 'Le Chonk,' claiming it ranks among the best open-weight models outside China. All three sources agree on the release and its focus areas: cyberdefense, coding, finance, manufacturing, and multimodal tasks.
There is some divergence on availability. TechCrunch says the model is not open-weight yet and is only accessible via a public guardrail endpoint, with weights planned in three weeks after safety testing. WIRED describes it as freely available in preview, with a final version by end of month. The distinction matters for enterprises deciding whether to adopt it.
Mistral positions Le Chonk as a third way between US closed models and Chinese open models, especially amid US restrictions on distributing models from OpenAI and Anthropic. The company says it trained the model from scratch on 4,000 Nvidia GPUs, two to three times fewer than Chinese competitors. Benchmarks are still pending, but Mistral says it is 'very, very close' to proprietary models in some areas.