OCR Reasoning on OCRBench v2
40.23AccuracyVLAA-Thinker-7B
Evaluation results
| Method | Links | |
|---|---|---|
| VLAA-Thinker-7B2025.10 | 40.23 | |
| RECAPBackbone=Qwen2.5-VL-7B, Training=finetuned2025.10 | 39.72 | |
| LwFBackbone=Qwen2.5-VL-7B, Training=finetuned2025.10 | 39.56 | |
| Qwen2.5-VL-7Bvariant=Base model2025.10 | 39.49 | |
| Reasoning-onlyBackbone=Qwen2.5-VL-7B, Training=finetuned2025.10 | 38.55 | |
| UniformBackbone=Qwen2.5-VL-7B, Training=finetuned2025.10 | 38.06 | |
| PropMixBackbone=Qwen2.5-VL-7B, Training=finetuned2025.10 | 37.6 | |
| CoresetBackbone=Qwen2.5-VL-7B, Training=finetuned2025.10 | 35.49 | |
| OpenVLThinker-7B2025.10 | 28.3 | |
| Vision-R1-7B2025.10 | 24.63 |