Mistral Large 4 Open Weights Released for Coding and Cyber Defense

Hakkı ALKAN

October 29, 2026

Mistral AI released the open weights of Mistral Large 4, its 1-trillion-parameter flagship model nicknamed “Le Chonk”, on October 29, 2026. Businesses and governments can now download the model from Hugging Face and run it on their own servers, while API access costs $0.70 per million input tokens and $2.10 per million output tokens.

The French company first showed Mistral Large 4 as a public preview earlier in October, as we covered in Mistral AI Unveils Mistral Large 4 with 1 Trillion Parameters. Mistral says three more weeks of reinforcement learning since then brought large gains in coding, defensive cybersecurity and reasoning efficiency.

Mistral Large 4 Specs: 1 Trillion Parameters, 1M Context

According to the official model card, Mistral Large 4 is a natively multimodal Mixture-of-Experts model with about 1 trillion total parameters and 49 billion active parameters per token (52 billion including embedding and output layers). Key details:

• 128 routed experts, 4 active per token, plus 1 shared expert
• 1 million token context window
• Text and image input, text output
• Support for more than 160 languages, including every official EU language
• Three reasoning modes: Low, Medium and High (default)
• Trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral’s own European data centers

The weights ship in FP8 and a smaller NVFP4 version aimed at Blackwell GPUs, with day-one support in transformers, vLLM, SGLang and llama.cpp, according to Hugging Face.

Coding and Cybersecurity Benchmarks

In its announcement, Mistral says the model scores 48% on the Artificial Analysis Coding Agent Index, effectively tied with China’s GLM-5.3, the strongest open-weight rival, while using 47% fewer tokens per task. On Terminal-Bench 4.0 it reached 41%, up 13 points from the preview, and it leads Cognition’s FrontierCode 2.0 benchmark with 19.7%.

For security work, Mistral reports a 51% score on the Artificial Analysis Cyber Index, which tests finding, reproducing and patching vulnerabilities. The company says built-in guardrails separate defensive tasks from offensive ones. Pierre Stock, Mistral’s VP of Science, told TechCrunch that open weights give organizations defensive AI access that “can’t be cut off”.

License, Price and Availability

Mistral Large 4 uses the Mistral Ecosystem Open-Weight License, a modified MIT license. Anyone can use, modify, fine-tune and sell products built on it, but companies offering it as a Model-as-a-Service need a separate commercial license. The model is also available in Mistral Studio, the Mistral API (mistral-large-4), OpenRouter and as an NVIDIA NIM container. Cached input costs $0.07 per million tokens.

What It Means

The release lands in a crowded week for frontier AI, days after Google’s Gemini 4 Carbon topped intelligence rankings. Closed labs like OpenAI and Anthropic are spending heavily on compute, as seen in Anthropic’s natural gas plans for its data centers. Mistral is betting that European governments and enterprises will prefer a top-tier model they can host and control themselves. A full technical report is coming soon, the company says.