Mistral’s Le Chonk: Can 4,000 GPUs Really Beat Big Tech?

According to ZDNET, French AI lab Mistral has officially released the model weights for its Mistral Large 4, widely known by its nickname “Le Chonk”. Built to tackle heavy development tasks, the model stands out by matching top proprietary competitors while running on significantly less computing infrastructure.

How does Le Chonk perform in everyday coding?

In benchmarks reported by ZDNET, Le Chonk jumped to a 41% score on the Terminal-Bench 4.0 agentic coding benchmark, up from its preview score of 28%. It now ranks alongside models like Codex GPT-6.1 Terra and rivals GLM-5.3 on the Artificial Analysis Coding Agent Index.

More importantly for daily workflows, the model achieves this performance while using 47% fewer tokens per task on average. For developers, fewer tokens mean faster response times and lower operational overhead when running automated coding agents locally or via API.

How was it built with limited compute?

While industry giants like OpenAI reportedly use up to 100,000 GPUs for flagship training runs, Mistral trained Le Chonk from scratch using just 4,000 Nvidia Grace Blackwell GPUs over two months. The training took place entirely within the lab's own European data centers, emphasizing data sovereignty and local control.

“The cyber defense capabilities will enable enterprises and governments to defend themselves against threat actors that are jailbreaking closed models to perform cyberattacks,” said Mistral co-founder Guillaume Lample.

Is it worth using for your development projects?

Le Chonk includes three distinct reasoning modes—low, medium, and high—giving developers flexibility depending on task complexity. Furthermore, its strong performance on the CyberGym benchmark makes it a practical option for teams looking to find and patch memory-safety vulnerabilities in C and C++ projects without relying on closed proprietary models.