Mistral announced today that a public preview of its new model, Mistral Large 4 (ML4), is available through the Mistral Studio API. The preview can be tried immediately, and the company plans to release the model weights by the end of the month. ML4 is a 1 trillion‑parameter model that uses 49 billion active parameters and is designed to handle text, code, and images.
How we got here
ML4 was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral’s European datacenters. The training data includes more than 160 languages, covering every official EU language. Mistral says the model already matches the strongest open‑source models on many benchmarks and beats other open‑weight models from the US and Europe, especially in cybersecurity, finance, and legal tasks.
What happens next
The company will continue red‑team testing with cybersecurity partners and state authorities before the final weights are released. After the weight release, customers will be able to run ML4 on private cloud or on‑premise installations, including a dedicated European deployment that operates under EU law. Mistral also plans to build specialized versions of ML4 for industry‑specific workloads, funded by its recent €3 billion Series D round.
Why it matters
For enterprises, especially in security and finance, ML4 offers a powerful AI that can be run on their own infrastructure, avoiding reliance on third‑party providers that might block critical tasks. The open‑weight release means organizations can audit and customize the model to meet regulatory or privacy requirements. However, the full capabilities and performance outside the preview remain unverified until the final weights are released.
