Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Mistral Large 4 joins Open Math OS with 1.05T open weights and 1M context
Mistral AI says Mistral Large 4 is now available on Open Math OS for approved teams, describing a 1.05T open-weight model with 1M context in a research workspace for mathematical work. The note is an official Mistral product tweet.
- Mistral Large 4 open weights land on Hugging Face
Hugging Face published the open weights of Mistral Large 4, codename Le Chonk: a ~1T-parameter Mixture-of-Experts model with 49B active parameters per token (52B including embeddings and output layers), a 1M-token context, text-and-image input, and fluency in 160+ languages. The release notes say it was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters. Weights ship under a MIT-derived Mistral Ecosystem Open-Weight License with FP8 and NVFP4 variants plus GGUF builds, and transformers v5.22.0 adds day-0 support alongside vLLM, llama.cpp, and TRL. Hugging Face highlights strong cybersecurity, coding-agent, and visual-grounding results, ranking it 5th among open-weight models on the Artificial Analysis intelligence index.
- Mistral launches Large 4 preview: 1T-parameter multimodal model, weights due end of month
Mistral launched a public preview of Mistral Large 4 (ML4), unofficially nicknamed Le Chonk: a 1 trillion-parameter, natively multimodal model with 49 billion active parameters. The preview API is live on Mistral Studio, with weights dropping at the end of this month. Mistral says ML4 is competitive with the strongest open-source models globally, significantly outperforms any open-weight model developed in the US or Europe, and is state-of-the-art among open models on enterprise workloads including cybersecurity, finance and law. In some domains such as visual grounding, the company says it surpasses even frontier closed models.
- Mistral shows off Mistral Large 4 on a 19-challenge CTF speedrun
Mistral posted benchmark-style results for Mistral Large 4, highlighting efficient reasoning over diverse complex challenges, including a CTF speedrun covering 19 challenges. The post is the company's first public performance demonstration since launch, as formal benchmark results remain pending.