Mistral is preparing to introduce a new model this summer and plans to open an early access program in July for a limited group of partners, according to company co-founder Arthur Mensch.
In a post on X, Mensch said the upcoming release will be the start of a new model family that is described as large but sparse. He said the model is expected to show notable improvements in some capabilities and suggested that Mistral wants it to be a surprise for users when it arrives.
The company said the early access program will target partners in research, government and industry. That approach suggests Mistral is aiming to give selected organizations a first look before a wider release. The company did not share a firm launch date for the model itself.
Mistral also said the new model, along with future releases in the family, will be open-weight. The company framed that choice as important for trust among customers and for the broader research and developer ecosystem. In its view, systems that can only be accessed through a third-party interface are harder to inspect, audit or improve.
The comments also highlighted Mistral’s deployment and training products, Studio and Forge. The company said both are portable and are now being hosted on infrastructure it controls. According to the post, those products can run in a customer’s virtual private cloud, in their own data center, or on Mistral infrastructure that is not tied to U.S. service providers.
Mistral said its computing capacity is already online and expanding quickly. It also said it can help customers secure that capacity. The company presented this infrastructure strategy as part of a broader effort to support organizations that want AI systems running outside of external control.
Forge, Mistral’s training tool, was described as a way to continuously train models using recorded human-AI interactions. The company said that capability is a key efficiency advantage for its approach.
Mensch’s remarks also placed Mistral’s work in a larger geopolitical and economic context. He argued that AI is becoming a major source of leverage and power, comparing its potential role to oil in the 20th century. He said the technology could either create broad prosperity or enable highly extractive economic systems.
Mistral positioned itself as pushing for the first outcome by advancing AI research and spreading access to the technology more widely. Mensch closed by noting that the company is hiring.
The post comes as Mistral continues to emphasize open-weight models and infrastructure independence as central parts of its product strategy. While the company offered few technical specifics, the announcement signals that another major release is on the way and that a small group of partners will get access before the broader public.