Short-form single-page QA on NewsQA (official)
39EMAGAR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AGARBackbone=Qwen3-VL-8B-Instruct2026.06 | 39 | 60.3 | 291 | |
| CGRBackbone=Qwen3-VL-8B-Instruct2026.06 | 38.5 | 59.3 | 268 | |
| RandomBackbone=Qwen3-VL-8B-Instruct2026.06 | 37.6 | 58.7 | 301 | |
| UniformBackbone=Qwen3-VL-8B-Instruct2026.06 | 37.6 | 58.7 | 292 | |
| AGLABackbone=Qwen3-VL-8B-Instruct2026.06 | 37.2 | 58.7 | 268 | |
| Plain VQABackbone=Qwen3-VL-8B-Instruct2026.06 | 36.7 | 58.3 | 268 | |
| VEA-ConBackbone=Qwen3-VL-8B-Instruct2026.06 | 36.4 | 58.1 | 268 | |
| VEA-BrBackbone=Qwen3-VL-8B-Instruct2026.06 | 36.2 | 57.7 | 268 | |
| VARBackbone=Qwen3-VL-8B-Instruct2026.06 | 13.5 | 30.9 | 268 |