Most AI labs name their flagship models like they're launching a luxury SUV. Mistral looked at its new trillion-parameter model and called it "le Chonk," and honestly, we respect a company that's comfortable with its size.
A Trillion Parameters, Made in Europe
Mistral AI opened a public preview of Mistral Large 4 on October 6. The company's own announcement says it's "unofficially ML4, very officially: le Chonk." It's a natively multimodal Mixture-of-Experts model with 1 trillion total parameters and 49 billion active at any given time, trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters, with training data covering more than 160 languages.
The preview API is available now on Mistral Studio at $1.36 per million input tokens and $4.18 per million output tokens, and open weights are due by the end of October. Mistral's benchmark claims include 61.7% on DeepSWE 1.1, 28.3% on Terminal Bench 4.0 and 93% on Cybench, which it describes as leading among open-weight models, plus a top-five spot on the AA Cyber Index.
Sovereignty You Can Download
The interesting part isn't the parameter count. It's the combination: frontier-scale model, European-built infrastructure, and weights you'll be able to run yourself. For organizations in regulated industries or the public sector, where "where does our data go?" ends every AI conversation, that's a real option rather than just a talking point. The 160-language training also matters for anyone serving audiences beyond English.
The angle people tend to miss: the strong cybersecurity scores are a double-edged sword once the weights are public. A model that's great at capture-the-flag challenges helps defenders find bugs faster, and it's equally available to anyone else who downloads it. Open weights put the responsibility for guardrails on whoever deploys the model, not on the lab. Also, "preview" means preview, so wait for independent benchmarks before you rebuild your stack around le Chonk.
Big model, big ambitions, and a nickname that will absolutely end up on a hoodie.
Weighing a self-hosted open model against an API for your business's AI features? We can help you sort out the trade-offs on cost, privacy and control before you pick a side.
Source: Mistral AI