Visual Question Answering on RIO-Bench Easy
75.81AccuracyQwen3-VL-8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3-VL-8BMethod Variant=Ours2026.07 | 75.81 | 162 | |
| Qwen3-VL-8BMethod Variant=Base2026.07 | 74.19 | — | |
| Qwen3-VL-4BMethod Variant=Ours2026.07 | 69.71 | 142 | |
| Qwen3-VL-30B-A3BMethod Variant=Ours2026.07 | 69.11 | 177 | |
| Qwen3-VL-4BMethod Variant=Base2026.07 | 68.29 | — | |
| Qwen3-VL-30B-A3BMethod Variant=Base2026.07 | 67.34 | — | |
| InternVL3.5-8BMethod Variant=Ours2026.07 | 63.96 | 9 | |
| InternVL3.5-8BMethod Variant=Base2026.07 | 63.87 | — | |
| InternVL3.5-14BMethod Variant=Ours2026.07 | 57.01 | 6 | |
| InternVL3.5-14BMethod Variant=Base2026.07 | 56.95 | — | |
| Gemma3-12BMethod Variant=Ours2026.07 | 52.27 | 236 | |
| Gemma3-12BMethod Variant=Base2026.07 | 49.91 | — | |
| Gemma3-4BMethod Variant=Ours2026.07 | 49.72 | 174 | |
| Gemma3-4BMethod Variant=Base2026.07 | 47.98 | — |