Mistral's New Model Has a Silly Name and a Serious Claim: It Beats China's Best

Follow Teck Hustlers on Google

Add Teck Hustlers as a preferred source to see more of our stories in Google Top Stories.

AA Admin Alex
Writer

MA M.Ali
Fact-Checker

TL;DR: Mistral released Mistral Large 4, nicknamed Le Chonk, a 1.05 trillion parameter open-weight model the French lab says outperforms the best Chinese open models on several benchmarks, putting real pressure on DeepSeek and Qwen’s claim to the open-source AI crown.

Somebody at Mistral clearly has a sense of humor, because the company’s biggest and most capable model to date is going out into the world nicknamed Le Chonk. Officially it’s Mistral Large 4, a 1.05 trillion parameter multimodal mixture-of-experts model now in public preview, and underneath the silly name is a genuinely serious claim: Mistral says it beats the leading Chinese open models on several benchmarks.

That’s not a throwaway line. For most of the last two years, the open-weight AI conversation has largely been a contest between Chinese labs, DeepSeek and Alibaba’s Qwen chief among them, both of which have repeatedly released models that matched or beat Western open alternatives while being free to download and run. Mistral, based in Paris, has been the most visible European answer to that pressure, and Le Chonk is its most direct swing yet at reclaiming that territory.

A mixture-of-experts architecture, for anyone who hasn’t had to care about this before, means the model isn’t actually running all 1.05 trillion parameters on every request. It routes each query through a smaller subset of specialized expert sub-networks instead, which is how modern frontier-scale models manage to be enormous on paper while staying affordable to run in practice. It’s the same basic approach behind most of the largest models shipping from any lab right now, just executed at a scale Mistral hadn’t attempted before.

Mistral’s own framing, echoed across early coverage from TechCrunch and CNBC, leans specifically on beating Chinese open models “in some areas,” a notably careful phrase. It’s not a claim of total dominance across every benchmark, and multimodal frontier models are notoriously difficult to compare cleanly anyway, since results shift depending on which tasks get tested. What it does signal is that Mistral believes it has closed, or at least narrowed, a gap that’s been a genuine sore spot for European and American open-weight AI for a while now.

The timing matters too. Open-weight models matter disproportionately to smaller companies, researchers, and governments that want frontier-level AI capability without being locked into a single closed API from OpenAI, Anthropic, or Google. A European lab shipping something genuinely competitive with the Chinese frontier gives that entire group a credible alternative that isn’t tied to either Beijing or Silicon Valley, which is as much a geopolitical talking point as a technical one right now.

Le Chonk is available in public preview starting today, with Mistral framing broader availability as the next step once initial feedback comes in. Expect independent benchmarking to follow quickly, since that’s the only way a claim like this actually gets tested outside a company’s own blog post.

Bottom Line: “Beats China’s best models in some areas” is a hedge, not a knockout claim, and that hedge is honestly more believable than a sweeping victory lap would have been. Mistral doesn’t need to win every benchmark to matter here. It just needed to show up credibly, and Le Chonk does that.