Document Derendering on DocHTML (test)
96.5Block AccuracyAnyDoc
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| AnyDocbackbone=Qwen2.5-VL-7B-Instruct2026.03 | 96.5 | 98.6 | 91 | 95.8 | 83.5 | 0.0625 | 0.0017 | 30.9 | |
| GPT-4omode=zero-shot2026.03 | 95.3 | 96.5 | 87.1 | 89.9 | 83 | 1.0009 | 0.0008 | 3.4 | |
| InternVL3-78Bmode=zero-shot2026.03 | 91 | 95.2 | 78.8 | 75.7 | 79.2 | 0.0298 | 0.0012 | 13.4 | |
| Design2Code2026.03 | 90.9 | 98.9 | 73.2 | 65.1 | 75.8 | 0.3756 | 0.001 | 65.2 | |
| WebSight2026.03 | 83.8 | 90.2 | 75.5 | 79.7 | 79.6 | 0.0873 | 0.0019 | 43.8 | |
| AnyDocBase Model=Qwen2.5-VL-7B-Instruct, Post-Training=HARL2026.03 | 0.965 | 0.986 | 0.91 | 0.958 | 0.835 | 0.0625 | 0.0017 | 0.309 | |
| GPT-4oModel Category=General MLLMs, Evaluation Setting=zero-shot2026.03 | 0.953 | 0.965 | 0.871 | 0.899 | 0.83 | 1.0009 | 0.0008 | 0.034 | |
| InternVL3-78BModel Category=General MLLMs, Evaluation Setting=zero-shot2026.03 | 0.91 | 0.952 | 0.788 | 0.757 | 0.792 | 0.0298 | 0.0012 | 0.134 | |
| Design2CodeModel Category=Specialized Models2026.03 | 0.909 | 0.989 | 0.732 | 0.651 | 0.758 | 0.3756 | 0.001 | 0.652 | |
| WebSightModel Category=Specialized Models2026.03 | 0.838 | 0.902 | 0.755 | 0.797 | 0.796 | 0.0873 | 0.0019 | 0.438 |