OCR VQA on InfoVQA (val)
81.4AccuracyQwen2.5-VL
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen2.5-VLOpenness=semi-open, Model Size=7B2025.10 | 81.4 | |
| Qwen3-VL-SegModel Size=4B, Training Stage=S-22026.05 | 81.4 | |
| Qwen3-VLModel Size=4B, Training Stage=instruct2026.05 | 80.3 | |
| InternVL-3.5Model Size=4B2026.05 | 78 | |
| InternVL2.5-MPOModel Size=8B2025.01 | 76.2 | |
| Qwen2-VLModel Size=8B2025.01 | 75.9 | |
| SAIL-VLModel Size=8B2025.01 | 75.2 | |
| Qwen3-VLModel Size=4B, Training Stage=S-12026.05 | 75.2 | |
| Bee-8BOpenness=fully open, Model Size=8B, Training Strategy=RL2025.10 | 72.9 | |
| DeepSeekVL-2Model Size=8B2025.01 | 72.5 | |
| Bee-8BOpenness=fully open, Model Size=8B, Training Strategy=SFT2025.10 | 72.3 | |
| Keye-VLOpenness=semi-open, Model Size=8B2025.10 | 67.4 | |
| SAIL-VLModel Size=2B2025.01 | 67.2 | |
| DeepSeekVL-2Model Size=2B2025.01 | 63.8 | |
| Qwen2-VLModel Size=2B2025.01 | 63.4 | |
| InternVL2.5-MPOModel Size=2B2025.01 | 61.6 |