Mistral previews 1T-parameter Large 4, weights drop in October

Mistral has released a public preview of Mistral Large 4, a 1 trillion-parameter mixture-of-experts model with 52 billion active parameters, and says the open weights will drop at the end of October.

"ML4 pushes the frontier of open-weight performance," the company writes, saying the model was "trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own datacenters in Europe." The nickname Mistral gives it: le Chonk.

The company's own reported benchmarks put ML4 at 61.7% on DeepSWE v1.1 and 28.3% on Terminal-Bench 4.0 for coding, and 42% on the Dense 200 visual-grounding test. On cybersecurity, Mistral says the model lands in the top five globally on the Artificial Analysis Cyber Index and "leads open-weight models developed outside China by a wide margin." None of these figures carries third-party replication at launch.

The release leans hard on the European-stack pitch. Mistral says the preview will be deployed across multiple regions worldwide, including an instance it operates "end-to-end, independently of other digital service providers and under European law." The company frames ML4 as the first milestone of a €3 billion Series D round, which it describes as "the largest equity round ever raised by a European technology company."

The preview API is live on Mistral Studio at $1.36 per million input tokens and $4.18 per million output tokens. Weights ship end of this month.