Visual Question Answering on RIO-Bench Medium
73.34AccuracyQwen3-VL-8B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3-VL-8BMethod Variant=Ours2026.07 | 73.34 | 2.11 | |
| Qwen3-VL-8BMethod Variant=Base2026.07 | 71.23 | — | |
| Qwen3-VL-30B-A3BMethod Variant=Ours2026.07 | 66.87 | 0.16 | |
| Qwen3-VL-30B-A3BMethod Variant=Base2026.07 | 66.71 | — | |
| Qwen3-VL-4BMethod Variant=Ours2026.07 | 66.52 | 0.85 | |
| Qwen3-VL-4BMethod Variant=Base2026.07 | 65.67 | — | |
| InternVL3.5-8BMethod Variant=Ours2026.07 | 61.18 | 0.41 | |
| InternVL3.5-8BMethod Variant=Base2026.07 | 60.77 | — | |
| InternVL3.5-14BMethod Variant=Base2026.07 | 55.65 | — | |
| InternVL3.5-14BMethod Variant=Ours2026.07 | 55.37 | -0.28 | |
| Gemma3-12BMethod Variant=Ours2026.07 | 49.84 | 1.1 | |
| Gemma3-12BMethod Variant=Base2026.07 | 48.74 | — | |
| Gemma3-4BMethod Variant=Ours2026.07 | 47.98 | 1.52 | |
| Gemma3-4BMethod Variant=Base2026.07 | 46.46 | — |