Visual Reasoning on CG-SalBench
66IoUQwen2.5-VL-72B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen2.5-VL-72BModel Category=Open-Source Model2026.04 | 66 | 82.7 | 82.37 | 86.74 | |
| ForeSight (7B)Model Category=Closed-Source Model2026.04 | 62.24 | 84.24 | 84.52 | 87.18 | |
| Qwen3-VL-8BModel Category=Open-Source Model2026.04 | 48.4 | 70.94 | 67.76 | 81.15 | |
| Doubao-Seed-1.6Model Category=Closed-Source Model2026.04 | 47.66 | 89.1 | 90.55 | 89.99 | |
| Qwen2.5-VL-7BModel Category=Open-Source Model2026.04 | 32.56 | 64.97 | 61.46 | 75.81 | |
| InternVL3-8BModel Category=Open-Source Model2026.04 | 30.5 | 72.09 | 69.74 | 79.01 | |
| Claude-Sonnet-4Model Category=Closed-Source Model2026.04 | 20.82 | 82.5 | 82.4 | 89.13 | |
| Gemini-2.5-FlashModel Category=Closed-Source Model2026.04 | 16.41 | 88.82 | 87.32 | 93.91 | |
| GPT-5Model Category=Closed-Source Model2026.04 | 7.85 | 89.85 | 87.79 | 95.54 |