Spatial Reasoning on CVBench
96.312D Relationship ScoreRoboRefer-8B-SFT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RoboRefer-8B-SFTModel Category=Spatial VLMs2025.12 | 96.31 | 96.5 | — | |
| GPT-5Model Category=Proprietary Models2025.12 | 95.54 | 91.33 | — | |
| Gemini-ER 1.5Model Category=Proprietary Models2025.12 | 95.54 | 90.5 | — | |
| SpaceTools-3BModel Category=Ours2025.12 | 94.92 | 96 | — | |
| Qwen2.5-VL-3B-Tool-freeModel Category=Tool-free Fine-tuning, Fine-tuning Protocol=SFT2025.12 | 91.54 | 83.33 | — | |
| Qwen2.5-VL-32BModel Category=General Open-Source Models2025.12 | 90.46 | 86.67 | — | |
| Claude Sonnet 4.5Model Category=Proprietary Models2025.12 | 89.85 | 78.5 | — | |
| GPT-4oModel Category=Proprietary Models2025.12 | 88.77 | 75.5 | — | |
| Qwen2.5-VL-3B-Tool-freeModel Category=Tool-free Fine-tuning, Fine-tuning Protocol=RL2025.12 | 87.38 | 70.83 | — | |
| RoboBrain2.0-7BModel Category=Spatial VLMs2025.12 | 87.23 | 90 | — | |
| RoboPoint-13BModel Category=Spatial VLMs2025.12 | 74 | 76.5 | — | |
| LLaVA-NeXT-8BModel Category=General Open-Source Models2025.12 | 72.15 | 73.67 | — | |
| Molmo-7BModel Category=Spatial VLMs2025.12 | 72.15 | 73.33 | — | |
| Qwen2.5-VL-3BModel Category=General Open-Source Models2025.12 | 70.62 | 65.33 | — | |
| SpaceLLaVA-13BModel Category=Spatial VLMs2025.12 | 61.08 | 62.83 | — | |
| Embodied-R1-3BAccess=Open-source2026.03 | — | — | 81.69 | |
| Gemini-3-flashAccess=Closed-source2026.03 | — | — | 83.94 | |
| GPT-5.2Access=Closed-source2026.03 | — | — | 79.62 | |
| PEEK-3BAccess=Open-source2026.03 | — | — | 55.49 | |
| Qwen3v1-4BAccess=Open-source2026.03 | — | — | 79.21 | |
| Robobrain2.0-3BAccess=Open-source2026.03 | — | — | 81.22 | |
| ST-VLAAccess=Ours2026.03 | — | — | 84.52 |