Embodied Referring on RefSpatial-Bench unseen
37.66AccuracyGeoThinker
Evaluation Results
| Method | Links | |
|---|---|---|
| GeoThinkerSFT Status=2B-SFT2026.02 | 37.66 | |
| RoboReferSFT Status=2B-SFT2026.02 | 33.77 | |
| Gemini-2.5-ProModel Category=Proprietary Models2026.02 | 27.14 | |
| Molmo-72BModel Category=Referring Specialist Models2026.02 | 21.24 | |
| Molmo-7BModel Category=Referring Specialist Models2026.02 | 12.23 | |
| RoboPointModel Category=Referring Specialist Models2026.02 | 8.4 | |
| SpaceLLaVAModel Category=Referring Specialist Models2026.02 | 4.02 |