Mistral has released the weights of Mistral Large 4, nicknamed “Le Chonk.” Anyone can now pull the European lab’s biggest model from Hugging Face and run it on their own hardware.
What’s confirmed. The model card lists about 1 trillion total parameters, with roughly 52 billion active per token once embeddings are counted. It reads text and images, supports a 1 million token context, and offers low, medium and high reasoning modes, with high as the default.
What Mistral claims. Mistral says it is the best open-weight model for coding, with 48% on the Artificial Analysis Coding Agent Index, roughly tied with GLM-5.3 while using 47% fewer tokens per task. It also claims the lead on Cognition’s private FrontierCode 2.0 test at 19.7%, which checks whether a maintainer would merge the agent’s code change.
On Terminal-Bench 4.0, Mistral reports a 13 point gain to 41% over the preview, with 30% fewer output tokens. It lists a 51% score on the Artificial Analysis Cyber Index and says the model refused 96.6% of malicious cyber requests on average, the best rate it measured among open models.
Its own card is more modest on general smarts: a score of 26 on the Artificial Analysis Intelligence Index, which it calls the top western open model, while admitting Chinese open models score higher.
Why now. Mistral put Large 4 in public preview earlier this month. It says about three more weeks of reinforcement learning since then drove the coding gains in this release, and that the model was trained from scratch on European infrastructure.
Read the license. It is a modified MIT license, which the card calls the Mistral Ecosystem Open-Weight License. You can use, tweak and sell products built on it, but companies that offer the model as a hosted service need a separate commercial deal with Mistral.
Our take. “Open” here mostly helps companies and governments that want to run AI on their own machines. At about 1 trillion parameters, this is not a laptop model. Even the smaller NVFP4 version listed on the model card is aimed at NVIDIA Blackwell GPUs, and Mistral says it trained the model on 3,800 Grace Blackwell GPUs in its own data centers in Europe.
What’s real here: a genuinely open frontier-scale model, mostly useful to people who own data center hardware.
Sources
- Mistral: Mistral Large 4 (Le Chonk) is now open weight
- Hugging Face: mistralai/Mistral-Large-4-1T-A52B
By Willa Ravenwire · The AI Bandwagon