Mistral’s new trillion-parameter model arrives as enterprises weigh not just AI performance, but also control, cost and dependence on proprietary platforms.

Who doesn’t love a chunky orange cat causing havoc? I certainly do. Millions of people agree that while they may not love AI, they like the French AI company Mistral‘s public preview of Mistral Large 4, nicknamed “Le Chonk.” This fat cat comes with a trillion-parameter open-weight large language model (LLM) that’s built for coding, cybersecurity, visual analysis, and enterprise agents.

Before diving into the tech side, Le Chonk has two things going for it that other AI models don’t. First, Mistral claims, with reason, to be the top sovereign AI. That makes it much more interesting to markets outside the US. Mistral has also made it clear that while Le Chonk will wander around the world, it will also offer a European deployment that the company will operate end-to-end, independently of other digital service providers and under European law.

I’m at Open Source Summit Europe in Prague, and everyone in Europe–Everyone–wants digital sovereignty. Governments and businesses that want to control their AI infrastructure are already finding Le Chonk more interesting than another chatbot claiming benchmark bragging rights. As one popular Reddit comment said, it’s a significant (and important) jump for a European model.

As Guillaume Lample, Mistral cofounder and chief scientist, told Wired, “Sometimes, people like to [make a big deal] over the US, versus Europe, versus China. But what really matters is to own the model—even for US companies,” says Lample. “If you use a closed model, there is no guarantee it will still be there tomorrow.”

Secondly, while its weights aren’t available yet, its license hasn’t been blessed as open source. Nevertheless, many people are getting excited about it because of its open-weights support alone. Indeed, we’re already seeing open-weight models grabbing more and more of the marketplace because they cost less than frontier models.

What about the Chinese open-weight models? In a Reuters interview, ​Mistral CEO Artur Mensch said, “The model we’re actually announcing today is actually above the Chinese models ⁠on certain aspects… the narrative that Europe cannot compete isn’t true.” ​

Moving to the tech side, Le Chonk uses a mixture-of-experts architecture, with 1 trillion total parameters and 49 billion active during inference. Its predecessor, Mistral Large 3, had 675 billion total parameters and 41 billion active.

Don’t confuse that active-parameter count with the model’s deployment footprint. A trillion-parameter model is not suddenly a 49-billion-parameter download. Mistral has yet to publish enough hardware and architectural detail to establish what practical self-hosting will cost.

The company says it trained ML4 from scratch on 3,800 Nvidia Grace Blackwell GPUs in its European data centers. The training data covers over 160 languages, including every official European Union (EU) language. That shows multilingual ambition, but it doesn’t prove equal proficiency across those languages.

ML4 also accepts multimodal input and produces text output. Mistral emphasizes understanding charts, PDFs, engineering drawings, and satellite imagery. It’s not about generating pictures. Mistral leaves that to Midjourney. No, its visual capabilities focus on helping agents locate relevant objects in images. So, for example, it could spot a particular component in a diagram or image to work with.

Mistral also claims Le Chonk is great at cybersecurity. On the Artificial Analysis Cyber Index, an independent evaluation of how well AI models find and fix security flaws in real software, the company states it “ranks among the top five models globally and leads open-weight models developed outside China by a wide margin. On one of the index’s tests, which asks a model to reproduce a real vulnerability in open-source software and then patch it, ML4 scores 82%, the highest of any model. It also solves 93% of the challenges in Cybench, a set of 40 exercises drawn from security competitions, one of the highest scores reported for an open-weight model.”

How well does it really work? We don’t really know yet. It’s only accessible by API. The version that everyone will be able to use won’t be out until late October. Still, the fat cat looks good in early looks.

Others are reporting on YComb everything from “Impressive vision benchmarking. If the vision model is truly as good as Astra, that would make it best in the world. Also strong on cyber benchmarks (better than all Chinese models), so this is a good defender model,” while others find it too slow.

“To sum it up, early testers see a meaningful improvement for Mistral, but reactions are divided over whether Le Chonk offers enough performance and value to displace established alternatives.” I’d argue, however, that what matters isn’t that ML4 is as fast as frontier models; it’s that it’s an open-weight model that enables anyone who values privacy and digital sovereignty to have an AI they can use and trust.