Models
Mistral's Robostral Navigate, an 8-billion-parameter model that steers robots with one RGB camera and a plain-language instruction, was announced on July 8, 2026, and its benchmark results come from Mistral
Mistral's Robostral Navigate, an 8B robot navigation model using one RGB camera, was announced July 8, 2026. Benchmark results are Mistral's own.
Mistral AI's first robotics model, Robostral Navigate, is an 8-billion-parameter model that takes images from a single RGB camera and a plain-language instruction and decides where a robot moves next. A correction to the framing that it is new this week: Mistral's own announcement thread and several outlets date it to July 8, 2026, and it is confirmed by Mistral rather than a single site. What remains vendor-reported is the performance: a 76.6% success rate on the R2R-CE benchmark, measured in simulation.
The essentials
- The model: 8 billion parameters, one ordinary RGB camera, a plain-language instruction and no LiDAR, depth sensor or pre-built map, per Mistral and press coverage.
- The benchmark: 76.6% success on the R2R-CE validation unseen set, which Mistral says beats the best single-camera approach by 9.7 points and the best depth or multi-camera system by 4.5 points. eWeek adds 79.4% on seen environments.
- Training: Mistral says it was trained entirely in simulation, on about 400,000 trajectories across 6,000 scenes, and runs on wheeled, legged and flying robots.
- What is unclear: real-robot results, availability, licence and price. One forum post says weights are public, which we could not corroborate.
Confirmed, but not new
The announcement dates to July 8, 2026, which Mistral's own post and outlets such as eWeek and MarkTechPost reflect. The recent coverage that prompted this item is a rehash of that launch, so the news value is the model itself and not a new release.
What the benchmark does and does not show
R2R-CE is a simulated instruction-following navigation test. A 76.6% success rate there says little about a physical deployment in a warehouse or factory. Claims that the model works across wheeled, legged and flying robots are Mistral's, and we saw no independent test.
What is not established
- Independent evaluation of the 76.6% result: not seen.
- Performance on real robots outside simulation: not shown.
- Open weights, licence and price: not confirmed.
- Customer deployments: none named.
Summary table
| Point | What is stated as of November 7, 2026 | Source |
|---|---|---|
| Model | Robostral Navigate, 8 billion parameters, Mistral AI | Pulse 2.0 |
| Inputs | One RGB camera plus a plain-language instruction | eWeek |
| Benchmark | 76.6% on R2R-CE validation unseen, per Mistral | MarkTechPost |
| Announced | July 8, 2026 | X (Mistral AI) |
| Training | In simulation, about 400,000 trajectories across 6,000 scenes | X (Mistral AI) |
FAQ
Is Robostral Navigate new?
No. Mistral announced it on July 8, 2026.
Does it need LiDAR?
No. Mistral says it uses one RGB camera and no depth sensor.
Have real robots been tested?
We saw no real-robot results; the benchmark is simulated.
Method and limits
Sources checked on November 7, 2026. We read the press coverage and saw Mistral's X thread only in search excerpts. Benchmark figures come from Mistral and are not independently verified. This is not investment advice.
Sources
- Pulse 2.0: Mistral AI Launches Robostral Navigate, An 8B Robotics Model That Steers Robots With A Single Camera (verified 2026-11-07)
- eWeek: Mistral AI's First Robot Model Navigates Using a Single Camera (verified 2026-11-07)
- MarkTechPost: Mistral AI Releases Robostral Navigate (verified 2026-11-07)
- X (Mistral AI): Mistral AI on X: Announcing Robostral Navigate (verified 2026-11-07)
- The AI Dude: Mistral Robostral Navigate: One Camera, 8B Params (verified 2026-11-07)
Spotted an error in this article? Report it · Our corrections policy