Tech & AI
Mistral opens the weights of its 1-trillion-parameter Large 4 model, pitching a European alternative to Chinese open models
The French lab says the model can run inside customers' own firewalls, with a licence that bars rivals from reselling it as a service.
HedgeWeekly Tech & AI Desk ·
Mistral AI, the Paris-based developer, released the open weights of Mistral Large 4 on October 29, about three weeks after a public preview. Open weights means the trained model files can be downloaded and run on a customer's own hardware. The model has 1 trillion parameters in total, with 52 billion active at a time, and is multimodal, meaning it handles more than text.
SiliconANGLE reported that Mistral trained it on 3,800 Nvidia Grace Blackwell chips in European data centres, and that chief executive Arthur Mensch says the company has raised roughly $6 billion in three years. Mistral lists API pricing of $0.70 per million input tokens and $2.10 per million output tokens, and says its own tests show the model matching GLM-5.3 on a coding-agent index while using 47% fewer tokens. Those benchmarks are the company's own.
The licence is a modified MIT licence: companies may use, modify and sell products built on it, but anyone offering the model itself as a paid service needs a separate commercial licence from Mistral. The weights are also offered through Nvidia, Microsoft Foundry and OpenRouter.
Why it matters: governments and banks wary of sending data to US clouds or Chinese models are the target market, and Mistral's route is to monetise through its own API and licences. Investors should watch independent benchmark results, uptake by regulated European customers, and whether Mistral's next funding round reflects this positioning.