Latest / Frontier models and safety
Mistral publishes the open weights of Large 4, its 1-trillion-parameter model
See how this story connectsMistral AI released the weights of Mistral Large 4 (Le Chonk), a multimodal mixture-of-experts model with about 1 trillion parameters, 49 billion of them active, and a 1-million-token context window, trained on European infrastructure. It is published under the Mistral Ecosystem Open-Weight licence, a modified MIT licence, in FP8 and NVFP4 versions. Mistral says it scores 19.7% on Cognition’s FrontierCode 2.0 benchmark and 51% on the Artificial Analysis Cyber Index; Hugging Face reports 82% on the index’s test of reproducing and patching a real vulnerability. It has three reasoning modes and refuses 96.6% of malicious cybersecurity requests on average, by Mistral’s measure.
Why it matters
Organisations can host and customise the open-weight model themselves for sovereign AI use, while Mistral limits preview access to its specialised cybersecurity variant to vetted enterprise design partners under contractual safeguards.
- When
- 29 Oct 2026
- ID
WTA-N-2026-0533- Where
- European Union, International
- Who
- Mistral AI, Booz Allen Hamilton, Cloudflare, CrowdStrike, Elastic, ENISA, Fortinet, NVIDIA, Orange Cyberdefense, Rubrik, Zscaler, La Banque Postale, Cognition, Artificial Analysis, Surge Code
Summary written with AI by wheresthe.ai, from Mistral AI and the reporting credited above. Checked against the source on 29 Oct 2026.
Source copy: saved on 29 Oct 2026 to check this story and not republished, because the text belongs to its publisher. Also held by the Internet Archive.
Cite this story
WTA-N-2026-0533: “Mistral publishes the open weights of Large 4, its 1-trillion-parameter model.” wheresthe.ai, 29 Oct 2026. https://wheresthe.ai/d/eu-mistral-releases-1-trillion-parameter-model-with-improved-2026-10/ (source: Mistral AI).
Spotted an error? Email [email protected].