Spatial Reasoning on CV-Bench 2D
94AccuracyQwen2.5-VL-3B
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-3BModel=Qwen2.5-VL-3B2026.03 | 94 | |
| 3ViewSense-4B-rl-slackTraining=RL, Reward=Slack2026.03 | 85.1 | |
| Qwen3-VL-8B-InstructModel=Qwen3-VL-8B-Instruct2026.03 | 84.7 | |
| 3ViewSense-4B-rl-strictTraining=RL, Reward=Strict2026.03 | 83.3 | |
| 3ViewSense-4B-sftTraining=SFT2026.03 | 80 | |
| SpaceQwen2.5-VL-3B-InstructModel=SpaceQwen2.5-VL-3B-Instruct2026.03 | 68.8 | |
| Qwen3-VL-4B-InstructModel=Qwen3-VL-4B-Instruct2026.03 | 68.8 | |
| SpatialLadder-3BModel=SpatialLadder-3B2026.03 | 68.6 | |
| InternVL3.5-4BModel=InternVL3.5-4B2026.03 | 66.1 | |
| SiTebackbone=LLaVA, variant=ratio2025.12 | 55.58 | |
| SiTebackbone=LLaVA-flickr, variant=ratio2025.12 | 55.07 | |
| SiTebackbone=LLaVA, variant=rand2025.12 | 54.28 | |
| SiTebackbone=LLaVA-flickr, variant=rand2025.12 | 54.06 | |
| LLaVAvariant=flickr2025.12 | 52.87 | |
| LLaVAvariant=Baseline2025.12 | 52.56 | |
| SiTebackbone=Qwen2, variant=rand2025.12 | 48.16 | |
| SiTebackbone=Qwen2, variant=ratio2025.12 | 47.47 | |
| Qwen2variant=Baseline2025.12 | 46.77 | |
| SiTebackbone=Qwen2-flickr, variant=ratio2025.12 | 44.44 | |
| SiTebackbone=Qwen2-flickr, variant=rand2025.12 | 43.64 | |
| SpaceOm-4BModel=SpaceOm-4B2026.03 | 43.1 | |
| Qwen2variant=flickr2025.12 | 42.59 |