Mistral Opens 1T-Parameter Le Chonk AI Model
Mistral AI has released the open weights of Mistral Large 4, its 1-trillion-parameter model nicknamed Le Chonk, free for anyone to download and run today.
Paris-based Mistral AI has released the open weights of Mistral Large 4, its flagship model nicknamed "Le Chonk", on October 29, 2026. Anyone can now download the 1-trillion-parameter model from Hugging Face and run it on their own servers, just three weeks after it first appeared in public preview.
What Mistral released
Mistral Large 4 is a natively multimodal mixture-of-experts model, which means it is built from many specialist sub-models and only activates the ones needed for each request. It has about 1 trillion parameters in total, with roughly 52 billion active per token (49 billion excluding embedding and output layers). It accepts text and image input, handles a 1-million-token context window, and is fluent in more than 160 languages. Mistral says the model was trained from scratch on 3,800 Nvidia Grace Blackwell GPUs in its own data centers in Europe. The company frames the release as a step for sovereign AI: organizations can run a frontier-class model inside their own firewalls, under European law, instead of relying on a US or Chinese provider.
Coding and cybersecurity gains
The three extra weeks of reinforcement learning since the October 6 preview brought clear gains, according to Mistral's own figures. On Cognition's private FrontierCode 2.0 benchmark, which tests whether a repository maintainer would merge an agent's code change, Mistral Large 4 leads open-weight models with a score of 19.7%. On the Artificial Analysis Coding Agent Index it scores 48%, effectively tied with China's GLM-5.3, while using 47% fewer tokens per task on average. On Terminal-Bench 4.0, a test of real terminal tasks, it improved 13 points from the preview to 41%, taking the lead among open-weight models while cutting output tokens by 30%. Mistral is also pitching the model to cyber defenders. It reports a 51% score on the Artificial Analysis Cyber Index, which measures finding, reproducing and patching vulnerabilities, and says the model refused 96.6% of malicious cyber requests in its tests. Three reasoning modes (low, medium and high) let developers trade performance against cost.
License, price and where to get it
The weights are published on Hugging Face under the Mistral Ecosystem Open-Weight License, a modified MIT license. Anyone can use, modify, fine-tune and commercialize the model, with one condition: businesses that offer it as a hosted model service need a separate commercial license from Mistral first. Through Mistral's own API, the model costs $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens. It is also available through OpenRouter, Nvidia-hosted endpoints with an optimized NIM container, and Microsoft Foundry, which Mistral says is the first cloud provider to carry it.
The safety debate
The release comes with an unusually open safety disclosure. Mistral says it monitored cybersecurity-related use in live traffic after the preview and saw the model refusing black-box attacks against live targets, but also saw "a potential for uplift to conduct real attacks". It says it is tightening its guardrails in response, with further updates promised in the coming months. A more aggressive variant built for penetration testing, vulnerability assessment and red-teaming is not being released openly. It stays restricted to vetted enterprise partners under contractual safeguards that limit testing to systems they own or are authorized to test.
What it means for open AI
Mistral Large 4 is now the highest-scoring Western open-weight model on the Artificial Analysis Intelligence Index, with a score of 26, though leading Chinese open-weight models still score higher on that aggregate index. Independent validation of Mistral's benchmark claims will take time, since the weights have only just landed. Still, the release matters because it closes much of the gap between open models and closed frontier models for coding work, and it does so with a license that lets companies actually deploy the model themselves. For teams that want control over their data and infrastructure, that combination is the point.
Sources
- Mistral AI - https://mistral.ai/news/mistral-large-4-oss/ - Unite.AI - https://www.unite.ai/mistrals-1-trillion-parameter-le-chonk-checkpoint-lands-on-hugging-face/