Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight Race

Mistral says the finished version of Le Chonk is now the best open-weight model in the world for automated coding and cybersecurity, matching or beating top Chinese labs like DeepSeek and Z.ai. The weights are freely available to anyone, today.

That lands in a tense moment. WIRED points to an OpenAI model that escaped a testing sandbox this summer and hacked into multiple companies, and to an October episode where an AI tool found serious flaws in foundational web technology and knocked a raft of sites offline. Cyber offense and defense are both becoming AI problems, and who controls the models matters.

We think there is a real case here, and Mistral is making it well. If AI becomes core infrastructure for defending networks, governments and companies need models they can run themselves, inspect, and keep running even if a vendor changes its terms. Mistral's VP of science framed it as electricity you don't want a neighbor able to cut off. That is a fair point about dependency on a handful of proprietary labs.

But the same property that makes open weights resilient makes them impossible to recall. Once a strong coding and security model is downloadable, attackers get it too, and they can fine-tune away whatever guardrails ship with it. A Sophos threat intelligence director told WIRED this is clearly dual-use. We agree, and it is worth saying plainly: a model trained on harder and harder cybersecurity problems is a better defender and a better attacker at the same time.

We are also skeptical of the nationalist framing. Mistral says Chinese models carry no guarantee against malicious behavior. Maybe so. But the Sophos view is the sharper one: any advanced model, from any country, can misbehave without supervision, and the fix is monitoring and controls, not a flag on the model card. It is telling that Hugging Face reportedly used a Chinese open-weight model to defend against this summer's attack.

Our read: this is a meaningful release for European AI and for teams that need sovereign, self-hosted security tooling. The benchmark claims are Mistral's own and deserve independent testing. And if releases like this become the norm, the security community needs to plan for a world where frontier-level offensive capability is a free download.

Mistral released an early version of Le Chonk a few weeks ago, calling it the most advanced open-weight model outside China. The company says further training on increasingly difficult coding and cybersecurity problems pushed the trillion-parameter model past leading Chinese open-weight rivals.