Mistral just dropped a 1 trillion-parameter AI model into the race.

Mistral just dropped a 1 trillion-parameter AI model into the race.

And it’s trying to prove Europe doesn’t need to choose between American closed models and Chinese open models.

Mistral Large 4, nicknamed “Le Chonk,” is a new multimodal model with 1 trillion parameters that Mistral says is designed to compete with both open and closed rivals.

Here’s the real story:

Mistral trained it using just 4,000 Nvidia GPUs — significantly fewer than the company says its Chinese and closed-model competitors typically use.

The model is also headed toward open weights. Mistral plans to release them in about three weeks, after additional safety testing and work with governments and trusted partners.

Its target markets are serious business: cybersecurity, finance, and chip design, with multimodal capabilities playing a major role.

The benchmark results aren't in yet, so the “leapfrog” claim remains a goal rather than a proven result.

But the strategy is fascinating.

Mistral isn't trying to be another OpenAI. It's trying to make Europe a third force in the AI model race.