Mistral AI released Robostral Navigate on July 8, 2026, the Paris-based company’s first model built specifically for physical robot navigation. CEO Arthur Mensch confirmed in the same week that a new “fat but sparse” mixture-of-experts model is entering partner early access as an open-weight release, according to Tech Insider.
How Robostral Navigate Works
The model is designed as hardware-agnostic: a single camera feed and a plain-language instruction are enough to move a robot through space it has never mapped before. Training was done entirely in simulation rather than on footage from physical robots, according to Mistral’s release notes cited by Tech Insider. If those claims hold under independent testing, robotics manufacturers could drop the model into existing sensor stacks without costly redesigns, a significant consideration for smaller manufacturers that cannot fund custom navigation software from scratch.
The timing is deliberate. Robostral Navigate shipped one day before OpenAI and xAI both released major updates of their own, positioning Mistral as a second major independent entrant in the physical AI space beyond the usual US robotics startups.
The Open-Weight Expansion
Alongside the robotics push, Mensch confirmed a new mixture-of-experts family described as “fat but sparse,” referring to a gating architecture where total parameter count is large but each token activates only a handful of specialized sub-networks. Mistral’s current flagship, Mistral Large 3, already runs this architecture with roughly 675 billion total parameters and about 41 billion active per token under an Apache 2.0 license, per Tech Insider.
The contrast with the rest of the industry continues to sharpen. OpenAI, Anthropic, and Google keep their most capable models fully closed. Mistral is betting that enterprises and governments increasingly want models they can inspect, self-host, and fine-tune rather than rent by the token.
Revenue and Valuation
Mistral’s annual recurring revenue passed $400 million in February 2026, up from roughly $20 million a year earlier, according to Tech Insider. Reports now put the company’s next funding round at more than $23 billion in valuation. For a company that started three years ago with a $260 million seed valuation, the trajectory is notable, though whether open weights can close the gap with proprietary rivals on raw benchmark performance remains the central open question.