Visual Reasoning on RCLIP-Bench Visual Reasoning V3
92.2Score (S)Human
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| HumanScale=Unbounded Scale2026.06 | 92.2 | 85.8 | 94 | 90.1 | 96.2 | 91.7 | |
| GPT-4VScale=Unbounded Scale2026.06 | 89.1 | 80 | 90.9 | 87.3 | 94.5 | 88.4 | |
| ReasonSigLIPScale=ViT So400m (428M), Resolution=384/14, Data=+42M2026.06 | 26.2 | 34.3 | 21 | 23.9 | 24.5 | 26 | |
| ReasonCLIPScale=ViT Large (307M), Resolution=224/14, Data=+42M2026.06 | 19.6 | 28.3 | 18 | 26.7 | 28.8 | 24.3 | |
| ReasonCLIPScale=ViT Base (86M), Resolution=224/32, Data=+42M2026.06 | 18.5 | 31.6 | 21.5 | 23.5 | 23.6 | 23.8 |