OCR and Chart Visual Question Answering on DocVQA (val)
95.5ScoreQwen2.5-VL
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VLOpenness=semi-open, Model Size=7B2025.10 | 95.5 | |
| Keye-VLOpenness=semi-open, Model Size=8B2025.10 | 88.5 | |
| Bee-8BOpenness=fully open, Model Size=8B, Training Strategy=SFT2025.10 | 87.2 | |
| Bee-8BOpenness=fully open, Model Size=8B, Training Strategy=RL2025.10 | 87 | |
| Emu3Param.=8B2026.02 | 76.3 | |
| HIVEBackbone=Huginn, Param.=4B, Train Data=6.5M, Recurrence Level (r)=322026.02 | 73.2 | |
| Gemma3-4B-PTBackbone=Gemma3, Param.=4B, Train Data=4T*2026.02 | 72.8 |