Mistral AI entered the next phase of the open-versus-closed AI contest on Tuesday with Mistral Large 4, a trillion-parameter multimodal model nicknamed “Le Chonk.” The French company says the release is meant to keep Europe relevant as Chinese developers dominate many open models and US groups such as OpenAI and Anthropic retain their closed-system lead.¹

The pitch is deliberately political as well as technical. ML4 is being framed as the “third way in AI” championed by French President Emmanuel Macron: a European alternative that can be self-hosted and modified, unlike closed US systems, without relying on Chinese technology.² Mistral said its model is particularly suited to cyber, coding, manufacturing, finance and multimodal work, and claims it will stand among the strongest open-weight systems once its core parameters are released.¹

For now, however, the model is not open. Mistral has put ML4 in preview through a guarded public endpoint for developers, cybersecurity leaders and state authorities, with a wider weight release planned later this month after safety testing. Pierre Stock, Mistral’s vice-president of science, said the company would work with “trusted partners and governments” to ensure the weights can be used “to defend, but not to perform malicious attacks.”²

That caution reflects a central contradiction in Mistral’s strategy. The company argues open weights make models easier to audit, yet acknowledges that powerful cyber capabilities demand a controlled rollout. Co-founder Guillaume Lample said those capabilities could help governments and businesses counter actors “jailbreaking closed models to perform cyber attacks.”¹

Mistral is also selling efficiency. It trained ML4 in its European data centres on 4,000 Nvidia Grace Blackwell GPUs over two months; Stock said that was “two to three times less than our Chinese competitors” and far less than closed-source rivals.² Whether that thrift translates into benchmark leadership remains the unanswered test of Europe’s bet.