Robostral Navigate
model Your tags
Your notes
The first model in Mistral's new Robostral line and its opening move in physical AI: an 8B vision-language model for robot navigation from a single RGB camera, RL-trained in simulation, reporting 79.4% on R2R-CE val-seen. Enterprise API only — no open weights or paper, a departure from Mistral's open-weight norm.
Model Details
Parameters 8B