Gemini Robotics-ER 1.6: benchmark results
Provider: Google. Access: API.
Aggregate Rating 1762 ± 38, rank #152 of 10916 rated models, from 53 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EgoSafetyBench | 88 | Video-level balanced accuracy (%; mean of the recall on unsa | 100 |
| RoboAbstention - Subjective Intent | 60.3 | Abstention rate (%) on the 242 Subjective Intent instruction | 100 |
| HELM Robo-Reward-Bench - Tokyo U Lsmo Converted Externally To Rlds | 0.49 | Absolute Error | 93.8 |
| WatchAct - Count | 40.6 | Plan success rate (%; the VLM's high-level Pick/Place/Open/C | 93.8 |
| HELM Robo-Reward-Bench - Droid | 0.6 | Absolute Error | 91.7 |
| EgoSafetyBench - Visual-Channel Mismatch Detection | 82.4 | F1 (%) of flagging chunks whose in-scene sign, sticker or la | 88.9 |
| HELM Robo-Reward-Bench - Austin Sirius Dataset Converted Externally To Rlds | 0.36 | Absolute Error | 87.5 |
| WatchAct - Conditional Execution | 39.8 | Plan success rate (%; the VLM's high-level Pick/Place/Open/C | 87.5 |
| WatchAct - Ordinal | 44.7 | Plan success rate (%; the VLM's high-level Pick/Place/Open/C | 87.5 |
| WatchAct - Task Continuation | 48.2 | Plan success rate (%; the VLM's high-level Pick/Place/Open/C | 87.5 |
| HELM Robo-Reward-Bench - Viola | 0.18 | Absolute Error | 83.3 |
| HELM Robo-Reward-Bench - Berkeley Rpt Converted Externally To Rlds | 0.63 | Absolute Error | 79.2 |
Interactive version: theaggregate.ai/model?slug=gemini-robotics-er-1-6 · How It Works · Data refreshed daily, snapshot 2026-11-08.