Short-form single-page QA on TriviaQA
64.7EMAGAR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AGARBackbone=Qwen3-VL-8B-Instruct2026.06 | 64.7 | 72.2 | 291 | |
| RandomBackbone=Qwen3-VL-8B-Instruct2026.06 | 63.2 | 70.4 | 301 | |
| AGLABackbone=Qwen3-VL-8B-Instruct2026.06 | 63.2 | 70.2 | 268 | |
| UniformBackbone=Qwen3-VL-8B-Instruct2026.06 | 62.6 | 69.6 | 292 | |
| Plain VQABackbone=Qwen3-VL-8B-Instruct2026.06 | 61.4 | 67.9 | 268 | |
| VEA-BrBackbone=Qwen3-VL-8B-Instruct2026.06 | 60.7 | 68.6 | 268 | |
| VEA-ConBackbone=Qwen3-VL-8B-Instruct2026.06 | 59.9 | 67.8 | 268 | |
| CGRBackbone=Qwen3-VL-8B-Instruct2026.06 | 51.5 | 67.3 | 268 | |
| VARBackbone=Qwen3-VL-8B-Instruct2026.06 | 46.9 | 63.6 | 268 |