OCR-related Understanding on DocVQA (val)
95.3AccuracyQwen3-VL
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen3-VLModel Size=4B, Training Stage=instruct2026.05 | 95.3 | |
| Qwen3-VL-SegModel Size=4B, Training Stage=S-22026.05 | 94.1 | |
| Qwen3-VLModel Size=4B, Training Stage=S-12026.05 | 93.8 | |
| InternVL-3.5Model Size=4B2026.05 | 92.4 |