Fine-grained Visual Reasoning on HR-Bench 8K
94.5FSP ScoreP2R-8B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| P2R-8BParameters=8B2026.07 | 94.5 | 70.8 | 82.6 | |
| P2R-4BParameters=4B2026.07 | 92 | 69 | 80.5 | |
| ZoomEye-7BParameters=7B2026.07 | 88.5 | 50 | 69.3 | |
| DeepEyes-7BParameters=7B2026.07 | 86.8 | 58.5 | 72.6 | |
| P2R-2BParameters=2B2026.07 | 86.8 | 62.7 | 74.8 | |
| Thyme-7BParameters=7B2026.07 | 86.5 | 57.5 | 72 | |
| PixelReasoner-7BParameters=7B2026.07 | 80 | 54.3 | 66.9 | |
| Qwen3-VL-Instruct-32BParameters=32B2026.07 | 78.8 | 68.5 | 73.6 | |
| Qwen3-VL-Instruct-4BParameters=4B2026.07 | 73.8 | 60.3 | 67 | |
| Qwen3-VL-Instruct-8BParameters=8B2026.07 | 73.3 | 70.1 | 70.1 | |
| Qwen3-VL-Instruct-2BParameters=2B2026.07 | 70.8 | 58.5 | 64.6 | |
| GPT-4o2026.07 | 62 | 49 | 55 |