Not established from the available sources.
MISTRAL AI · MODEL RELEASE TRACKER
Robostral Navigate
Robostral Navigate is an 8B embodied navigation model introduced by Mistral AI on 2026-07-08. It uses a single RGB camera and plain-language instructions to autonomously navigate robots through complex environments, was trained entirely in simulation, and reports 76.6% success on R2R-CE validation unseen.CURRENT SNAPSHOT2/5 DIMENSIONS WITH DATA
The dimensions that change the decision.
Not established from the available sources.
Input modalitiesOperates from a single RGB camera (RGB images) and plain‑language instructions; does not use LiDAR or depth sensors.
R2R-CE benchmark performance76.6% success on R2R-CE validation unseen; 79.4% success on validation seen. Reported to beat the best single-camera approach by 9.7 points and the best multi-sensor/depth system by 4.5 points.
Not established from the available sources.
VERIFIABLE FACTS
Every value stays attached to a source and date.
RELEASE · Announcement dateDEVELOPER CLAIM
Introduced on 2026-07-08 (publication: Mistral AI)
CAPABILITIES · Model sizeDEVELOPER CLAIM
8B parameters
MODALITIES · Input modalitiesDEVELOPER CLAIM
Operates from a single RGB camera (RGB images) and plain‑language instructions; does not use LiDAR or depth sensors.
BENCHMARKS · R2R-CE benchmark performanceDEVELOPER CLAIM
76.6% success on R2R-CE validation unseen; 79.4% success on validation seen. Reported to beat the best single-camera approach by 9.7 points and the best multi-sensor/depth system by 4.5 points.
CAPABILITIES · Training and methodsDEVELOPER CLAIM
Built entirely in-house and trained entirely in simulation; combines pointing-based navigation with reinforcement learning for continuous improvement.
CAPABILITIES · Generalization across robot typesDEVELOPER CLAIM
Runs on wheeled, legged, and flying robots and generalizes across robot sizes; claims to adapt to real-world obstacles unseen during training.
WHAT CHANGED