High-resolution Visual Question Answering on HR-Bench 4K
82.1Overall AccuracyQwen3-VL-32B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-VL-32B2026.05 | 82.1 | 94 | 70.2 | |
| PFlowNet2026.05 | 80.4 | 91.2 | 69.5 | |
| Qwen2.5-VL-72B2026.05 | 79.4 | 88.8 | 70 | |
| DeepEyesV22026.05 | 77.9 | 92.8 | 63 | |
| TreeVGR2026.05 | 77.1 | 89.5 | 64.8 | |
| Thyme2026.05 | 77 | 91 | 63 | |
| InternVL3-38B2026.05 | 76.3 | 83.5 | 69 | |
| DeepEyes2026.05 | 75.1 | 91.3 | 59 | |
| Qwen2.5-VL-32B2026.05 | 74.8 | 89.3 | 60.3 | |
| Qwen3-VL-4B2026.05 | 73.5 | 84.8 | 62.3 | |
| PixelReasoner2026.05 | 72.9 | 86 | 60.3 | |
| Qwen3-VL-8B2026.05 | 72.4 | 88.5 | 56.3 | |
| Qwen2.5-VL-7B2026.05 | 72.1 | 88.8 | 55.5 | |
| InternVL3-8B2026.05 | 70.8 | 79.3 | 62.3 | |
| LLaVA-OV-72B2026.05 | 66.3 | 76.5 | 56 | |
| GPT-4o-11202026.05 | 59 | 70 | 48 |