Mistral news and engineering summaries

Mistral Large 4 is a state-of-the-art, open-weight, general-purpose multimodal model released in Public Preview v26.10 on October 6, 2026. Its granular Mixture-of-Experts architecture has 49B active parameters and 1.05T total parameters, alongside a 1.6B vision encoder. The model supports Structured Outputs, Function Calling, Document QnA, Chat Completions, batching, Agents & Conversations, Built-In Tools, and Prefix features.

Mistral launched a public preview of Mistral Large 4 (ML4), a 1-trillion-parameter natively multimodal MoE with 49 billion active parameters. Trained on 3,800 NVIDIA Grace Blackwell GPUs in European datacenters, it targets coding, cybersecurity, agentic workflows, science, finance, law, and visual grounding. ML4 scores 82% on a CyberGym vulnerability task, 93% on Cybench, and 49.8% on the Coding Agent Index. The preview API is available on Mistral Studio; weights are planned by month-end.

Mozilla and Mistral are partnering to power Firefox Smart Window beta with Mistral models, initially for users in France and North America, with the United Kingdom and Germany planned next. The collaboration combines open models, regional language adaptation, and Firefox’s privacy-first design, including default non-storage of conversations and zero data retention, to give users greater control over AI-powered browsing.

Mistral develops general-purpose and specialist AI models for reasoning, language, multimodal understanding, coding, document intelligence, audio, robotics, and physical AI. Its publications include LLaMa, Chinchilla, Voxtral, Mixtral of Experts, Mistral 7B, and Pixtral 12B, covering scaling, sparse MoE, efficiency, and image or document understanding. Mistral also invites users to download, adapt, and deploy its open-weight models and highlights research spanning training, evaluation, inference, and efficiency.