Grounding Accuracy on DocVQA BBox
100Good RatioQwen3VL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen3VLparameters=235B, variant=a22b-thinking2026.06 | 100 | 52.11 | 45.93 | 42.09 | 48.03 | |
| PreciseDoc-Reasoner2026.06 | 100 | 58.93 | 41.01 | 50.61 | 51.16 | |
| GPT-52026.06 | 99.9 | 0.9 | 0.1 | 1.6 | 1.2 | |
| GPT-5.22026.06 | 99.6 | 0.88 | 3.31 | 3.39 | 2.21 | |
| Gemini-3variant=Pro2026.06 | 99.6 | 54.7 | 45.18 | 37.87 | 48.15 | |
| Gemini-3variant=Flash2026.06 | 99.4 | 63.56 | 56.42 | 59.11 | 60.24 | |
| Qwen2.5VLparameters=72B2026.06 | 99 | 40.1 | 33.2 | 27.2 | 35.2 | |
| InternVL3parameters=8B2026.06 | 98 | 0.1 | 0.5 | 0.5 | 0.3 | |
| Doubao-Seed-1-82026.06 | 97.2 | 55.07 | 47.91 | 37.85 | 49.25 | |
| Qwen3VLparameters=4B2026.06 | 95.9 | 18.9 | 17 | 21.3 | 18.7 | |
| Qwen3VLparameters=32B2026.06 | 94.5 | 22.6 | 17.3 | 21 | 20.4 | |
| Qwen2.5VLparameters=32B2026.06 | 94 | 22.2 | 20.4 | 14.3 | 20 | |
| Qwen3.5parameters=27B2026.06 | 93.5 | 56.27 | 46.01 | 17.2 | 44.5 | |
| Qwen3VLparameters=8B2026.06 | 92.7 | 17.6 | 9.2 | 15.9 | 14.4 | |
| InternVL3parameters=2B2026.06 | 80.9 | 0 | 0.2 | 0.1 | 0.1 | |
| Qwen2.5VLparameters=7B2026.06 | 73 | 14.6 | 8.8 | 8 | 11.3 | |
| Qwen2.5VLparameters=3B2026.06 | 51.4 | 3.8 | 5.6 | 4.9 | 4.7 |