Fine-grained Visual Reasoning on HR-Bench-8K OOD
88.5FSP ScoreDeepEyesV2
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DeepEyesV22026.07 | 88.5 | 59 | 73.8 | |
| DeepEyes2026.07 | 86.8 | 58.5 | 72.6 | |
| Thyme2026.07 | 86.5 | 57.5 | 72 | |
| TreeVGR2026.07 | 86 | 59.5 | 72.8 | |
| TikArtBase Model=Qwen3-VL-8B2026.07 | 84.5 | 68.3 | 76.4 | |
| Gemini-2.5-pro2026.07 | 84.5 | 82.8 | 83.6 | |
| Qwen2.5-VL-7B2026.07 | 83.5 | 54 | 68.8 | |
| Qwen3-VL-4B2026.07 | 83.5 | 50.7 | 67.1 | |
| SIEVEBase Model=Qwen3-VL-8B2026.07 | 83 | 73.5 | 78.3 | |
| BUS-GRPOBase Model=Qwen3-VL-8B2026.07 | 82.3 | 70.8 | 76.5 | |
| BUS-SFTBase Model=Qwen3-VL-8B2026.07 | 81.5 | 56.3 | 68.9 | |
| Qwen3-VL-8B2026.07 | 81.3 | 55.8 | 68.5 | |
| PixelReasoner2026.07 | 80 | 54.3 | 66.9 | |
| Gemini-2.5-flash2026.07 | 79 | 73.5 | 76.3 | |
| V-Reflection2026.07 | 73.5 | 58.5 | 66.3 | |
| GPT-52026.07 | 70.5 | 72.8 | 71.6 | |
| GPT-5-nano2026.07 | 65.8 | 61.5 | 63.6 | |
| InternVL3-8B2026.07 | 64.3 | 59.8 | 62 | |
| GPT-4o2026.07 | 60.8 | 60 | 60.4 |