Visual Spatial Perception on VSP Seen
95Accuracy (Level 3)Gemini-2.5-flash
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Gemini-2.5-flash2026.05 | 95 | 86 | 62 | 30 | 62 | |
| CoLVR2026.05 | 94 | 92 | 80 | 57 | 80.7 | |
| Qwen2.5-VL-7BTraining Paradigm=vanilla SFT2026.05 | 91 | 85 | 71 | 50 | 74.3 | |
| Mirage2026.05 | 90 | 84 | 67 | 55 | 74 | |
| Qwen2.5-VL-7BTraining Paradigm=vanilla SFT + GRPO2026.05 | 89 | 80 | 66 | 49 | 71 | |
| GPT-4o2026.05 | 67 | 46 | 24 | 17 | 38.5 | |
| InternVL3.5-8B2026.05 | 20 | 13 | 3 | 2 | 9.5 | |
| Qwen2.5-VL-7BReasoning Strategy=CoT2026.05 | 20 | 8 | 5 | 3 | 9 | |
| Qwen2.5-VL-7B2026.05 | 11 | 7 | 7 | 1 | 6.5 |