OCR-related understanding on DocVQA
95.1ScoreOvis2.5
Evaluation Results
| Method | Links | |
|---|---|---|
| Ovis2.5Parameter Scale=9B2026.01 | 95.1 | |
| Qwen3-VLParameter Scale=8B2026.01 | 94.4 | |
| 1.7B K^8_{0.5}Size=18B-A3B, Sparsity Configuration=K=8, alpha=0.52026.01 | 93.8 | |
| Qwen2.5-VLParameter Scale=72B2026.01 | 93.3 | |
| 1.7B K^4_{1.0}Size=18B-A3B, Sparsity Configuration=K=4, alpha=1.02026.01 | 92.9 | |
| InternVL3.5-20B-A4BSize=20B-A4B2026.01 | 92.9 | |
| Isaac 0.2Size=2B2026.01 | 92.1 | |
| GLM-4.6V-FlashX2026.01 | 92 | |
| LoMoModel=Qwen3.5-9B, Evaluation Protocol=Standard Evaluation2026.05 | 91.99 | |
| LoMoModel=Qwen3.5-9B, Evaluation Protocol=Rendered Evaluation2026.05 | 91.73 | |
| InternVL3.5Parameter Scale=8B2026.01 | 90.6 | |
| Standard SFTModel=Qwen3.5-9B, Evaluation Protocol=Standard Evaluation2026.05 | 85.89 | |
| Ostrakon-VLParameter Scale=8B2026.01 | 85.7 | |
| LoMoModel=LLaVA-OV1.5-8B, Evaluation Protocol=Standard Evaluation2026.05 | 74.77 | |
| Standard SFTModel=LLaVA-OV1.5-8B, Evaluation Protocol=Standard Evaluation2026.05 | 73.05 | |
| LoMoModel=LLaVA-OV1.5-8B, Evaluation Protocol=Rendered Evaluation2026.05 | 58.89 | |
| GenLIPArch=g/16, Data=8.0B2026.05 | 57 | |
| GenLIPArch=So/16, Data=8.0B2026.05 | 51.9 | |
| GenLIPArch=L/16, Data=8.0B2026.05 | 51.1 | |
| SigLIP2Arch=g/16, Data=40.0B2026.05 | 47.6 | |
| SigLIPArch=L/16, Data=40.0B2026.05 | 47.3 | |
| Standard SFTModel=Qwen3.5-9B, Evaluation Protocol=Rendered Evaluation2026.05 | 47.24 | |
| SigLIP2Arch=So/16, Data=40.0B2026.05 | 46.4 | |
| SigLIP2Arch=L/16, Data=40.0B2026.05 | 45.1 | |
| OVision2Arch=L/16, Data=12.8B2026.05 | 43.3 | |
| CLIPArch=L/14, Data=12.8B2026.05 | 38.9 | |
| AIMv2Arch=L/14, Data=12.0B2026.05 | 37.7 | |
| Standard SFTModel=LLaVA-OV1.5-8B, Evaluation Protocol=Rendered Evaluation2026.05 | 15.38 |