On Tuesday, French AI lab Mistral unveiled Mistral Large 4, or “Le Chonk,” a natively multimodal model with roughly 1 trillion parameters and 49 billion active parameters. The company’s pitch is deliberately geopolitical as well as technical: an open-weight European option between US proprietary systems and the increasingly influential Chinese open-model ecosystem.¹

Mistral says the model was trained from scratch on 4,000 Nvidia Grace Blackwell GPUs in its European data centres. Chief executive Arthur Mensch emphasized the independence behind the launch, writing that it was “trained and served on our own compute,” with reinforcement learning showing “no sign of saturation.”Arthur Mensch (arthurmensch)² The company argues ML4 can excel in cyber defence, coding, manufacturing and finance—specialist work where it believes focused training can matter more than broad, headline-grabbing reasoning scores.

That ambition arrives after Mistral’s September €3 billion Series D, which valued the company at €21 billion. Co-founder Guillaume Lample framed the goal bluntly: “Mistral is still in the race of getting the best model.” He also made the company’s central case for open weights: “If you use a closed model, there is no guarantee it will still be there tomorrow.”³

For now, though, Le Chonk is only in preview behind a guardrailed endpoint. Mistral plans to release its weights after roughly three weeks of safety testing, working first with trusted partners and governments to ensure the model can “defend, but not … perform malicious attacks,” VP of Science Pierre Stock said.⁴

The gap between promise and proof remains material. One independent comparison places ML4 behind leading proprietary frontier models and Chinese open-weight rivals on general intelligence, even as it identifies strong speed and specialised agentic performance. In other words, Mistral may have produced the West’s strongest open-weight challenger—yet its larger sovereignty bet depends on benchmarks, audits and safe deployment catching up with the hype.⁵