Mistral AI Launches Robostral Navigate, Its First Robot Navigation Model

Mistral AI has introduced Robostral Navigate, an 8-billion-parameter model built for autonomous robot navigation in real-world indoor environments. The model operates using only a single standard RGB camera and natural-language instructions, eliminating the need for expensive sensors like LiDAR or depth cameras. Trained entirely in simulation on roughly 400,000 navigation trajectories, it achieved a 76.6% success rate on the R2R-CE benchmark, outperforming other single-camera approaches. The hardware-agnostic model supports wheeled, legged, and flying robots, potentially lowering the cost barrier for deploying autonomous systems. Mistral has not announced pricing or a commercial release date, but is actively expanding its robotics team, signaling a long-term commitment to physical AI.