Mistral Large 4 (ML4) enters a landscape where large language models are increasingly divided between proprietary systems that can be remotely disabled and open-weight models, many of which originate in China. Mistral is positioning Le Chonk as an alternative that combines openness with safety assurances.
For now, ML4 can only be accessed via a public guardrail endpoint. "In the meantime, we'll work with trusted partners and governments to make sure that the open-source weights can be used to defend, but not to [perform] malicious attacks," Mistral VP Science Pierre Stock told TechCrunch. Stock noted that an open-weight model is also easier to audit, addressing security concerns among Mistral's enterprise and institutional customers.
A notable technical claim is training efficiency: ML4 was trained on only 4,000 NVIDIA GPUs using Mistral's own compute. Stock said this is "two to three times less than our Chinese competitors, and significantly less than the closed source competitors."
Benchmark results are pending, but Mistral expects ML4 to be best in class among open-weight models outside China, and potentially to outperform closed models in targeted areas. The company says the model is optimized for cybersecurity, finance, and chip design. The latter is a strategic area given Mistral's key backers: ASML, which led its Series C, and Samsung, which led a Series D last month at a €21 billion valuation.
The release comes amid French President Emmanuel Macron's public support for what he has called "a third way in AI." Mistral is framing ML4 as evidence it remains a frontier lab rather than merely an inference provider, following its recent decision to host Chinese models.
