Visual Spatial Perception on VSP Unseen
43Accuracy (Level 7)CoLVR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CoLVR2026.05 | 43 | 29 | 36 | |
| Mirage2026.05 | 39 | 25 | 32 | |
| Qwen2.5-VL-7BTraining Paradigm=vanilla SFT2026.05 | 36 | 29 | 29 | |
| Qwen2.5-VL-7BTraining Paradigm=vanilla SFT + GRPO2026.05 | 34 | 14 | 24 | |
| Gemini-2.5-flash2026.05 | 32 | 18 | 25 | |
| GPT-4o2026.05 | 11 | 6 | 8.5 | |
| Qwen2.5-VL-7BReasoning Strategy=CoT2026.05 | 3 | 2 | 2.5 | |
| InternVL3.5-8B2026.05 | 2 | 0 | 1 | |
| Qwen2.5-VL-7B2026.05 | 1 | 0 | 0.5 |