Mistral AI, the Paris-based artificial intelligence company that has positioned itself as Europe's answer to OpenAI, has launched a public preview of Mistral Large 4, its biggest and most capable model to date.
Known internally as "le Chonk," the model packs one trillion parameters in total with 49 billion active at any one time, making it a mixture-of-experts architecture: a design that activates only a portion of the model for each task to keep costs and latency manageable.
What it can do
The headline claim is cybersecurity performance.
On the Artificial Analysis Cyber Index, an independent benchmark measuring how well AI models find and fix real software vulnerabilities, Mistral Large 4 ranks in the global top five and leads all open-weight models built outside China by a significant margin.
Closed rivals including Claude Opus 5.5 and GPT-6 Astra score near zero on one of the index's key tests because their safety filters refuse to perform the task, which involves reproducing a known vulnerability and then patching it: Mistral Large 4 scores 82%.
Beyond security, the model posts competitive results on agentic coding benchmarks and business workflow tasks, and Mistral claims it surpasses GPT-6 Astra on both legal and financial task evaluations conducted by third-party assessors.
Built in Europe
The model was trained on 3,800 Nvidia Grace Blackwell GPUs in Mistral's own European datacentres, and the company says it will remain available under a European deployment independent of US cloud infrastructure.
The full model weights are due at the end of October, at which point Mistral says it will publish full architectural details and benchmark methodology.
