Mistral has released the open weights of Large 4, nicknamed “Le Chonk”, its 1-trillion-parameter natively multimodal model — an upgrade from the public preview unveiled on October 6, after three further weeks of reinforcement learning focused on coding and cybersecurity. The weights landed on Hugging Face on October 29, 2026, with coverage following on October 31, and anyone can now download and run the model on their own infrastructure with tools such as vLLM or SGLang, rather than calling it only through Mistral’s API.

Mistral’s own figures put the model at the top of open-weight coding: 19.7% on FrontierCode 2.0, 41% on Terminal-Bench 4.0 (up from 28% in the preview), and 51% on the Artificial Analysis Cyber Index, with 52 billion active parameters per token. Those are company numbers, not independently verified results, and the model is aimed squarely at organisations that want to find and fix security vulnerabilities on systems they control. Reuters reports Mistral positioning it as its most capable openly available model for cybersecurity use, in a field where closed models still dominate.

Two catches deserve equal billing. First, the download is enormous — reporting puts the smallest official version at about 610GB, with a full-precision version around 1.06TB — so this is datacentre hardware, not a laptop model. Second, “open weights” is not full open source: Mistral has not published its training data, and businesses that want to offer the model as a service must obtain a separate commercial licence from Mistral. For defenders, that trade is still attractive: full control over deployment, with no third-party provider in the loop.

Sources