French AI lab Mistral AI has released Mistral Large 4 (ML4), a new large multimodal model positioned as a strategic alternative to both American closed models and Chinese open-weight rivals. The company describes the release as part of a "third way" in AI, a phrase echoed by French President Emmanuel Macron.

What Happened

Nicknamed "Le Chonk" in reference to its 1 trillion parameters, ML4 is currently accessible only via a public guardrail endpoint. Mistral plans to release the model's weights in three weeks, following the completion of safety testing. Pierre Stock, Mistral’s VP of Science, stated that the interim period will be used to work with trusted partners and governments to ensure the open-source weights are utilized for defensive purposes rather than malicious attacks.

Stock noted that while security concerns have mounted among Mistral’s core enterprise and institutional audience, an open-weight model remains easier to audit. The company reports that ML4 was trained entirely on Mistral’s own compute infrastructure, utilizing 4,000 Nvidia GPUs. According to Stock, this represents two to three times less compute than Chinese competitors and significantly less than closed-source rivals.

Why It Matters

Mistral’s strategy highlights the growing geopolitical and technical divide in the AI landscape, where open models are frequently associated with Chinese development and closed models with American labs. By aiming to be "best in class" among open-weight models outside of China, and potentially outperforming closed models in specific areas, Mistral is attempting to carve out a distinct niche for European sovereign AI.

The company identifies cybersecurity, finance, and chip design as key optimized use cases for ML4. These sectors align with the interests of Mistral’s major backers, including ASML, which led the company’s Series C, and Samsung, which led its Series D last month at a €21 billion valuation. Despite previously hosting Chinese models, Mistral asserts that ML4 reinforces its status as a frontier lab rather than merely an inference provider.

The Bottom Line

With benchmark results still pending, Mistral is betting that efficient training on a smaller GPU cluster and a delayed but secure open-weight release will allow ML4 to compete with larger, resource-intensive rivals from both the US and China.