OCR Performance Evaluation on OCRBench
94ScoreGemini 3.0 Pro
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Gemini 3.0 ProSize=Closed2026.03 | 94 | — | |
| Qwen3-VLSize=235B(A22B)2026.03 | 92 | — | |
| InternVL3.5Size=78B2026.03 | 90.6 | — | |
| Qwen3-VLSize=8B2026.03 | 89.6 | — | |
| Qwen3-VLSize=8B, mode=instruct2026.04 | 89.6 | — | |
| InternVL3.5-38BModel Scale=Large, Framework=Baseline, Visual Highlighting=false2026.04 | 89 | — | |
| Qwen2.5-VLSize=72B2026.03 | 88.5 | — | |
| Qwen3-VLSize=4B2026.03 | 88.1 | — | |
| Qwen3-OmniSize=30B-A3B, mode=instruct2026.04 | 88 | — | |
| MiniCPM-o 4.5Size=9B, mode=instruct2026.04 | 87.6 | — | |
| GLM-4.5VSize=106B(A12B)2026.03 | 87.2 | — | |
| InternVL3.5Size=38B2026.03 | 87 | — | |
| InternVL3.5Size=20B2026.03 | 87 | — | |
| Qwen2.5-VLSize=7B2026.03 | 86.4 | — | |
| Gemini 2.5 FlashSize=-, mode=instruct2026.04 | 86.4 | — | |
| InternVL3.5-38B + LoTModel Scale=Large, Framework=LoT, Visual Highlighting=true2026.04 | 86.2 | — | |
| TokenVL-8BSize=8B2026.03 | 86 | — | |
| HunyuanOCRSize=0.9B2026.03 | 86 | — | |
| Qwen3-VLSize=2B2026.03 | 85.8 | — | |
| Q-MaskSize=3B2026.03 | 85.5 | — | |
| InternVL3-8B2025.10 | 85 | — | |
| InternVL3-8B + CARESCost=-70%2025.10 | 85 | — | |
| InternVL3-8B + CARES-ARCost=-78%2025.10 | 85 | — | |
| Qwen2.5-VL-7BModel Scale=Medium, Framework=Baseline, Visual Highlighting=false2026.04 | 84.9 | — | |
| Qwen2.5-VL-7BBase Model=Qwen2.5-VL-7B2026.06 | 84.4 | — | |
| Qwen2.5-VL-7B + LoTModel Scale=Medium, Framework=LoT, Visual Highlighting=true2026.04 | 84.3 | — | |
| Qwen3-VL-8B + LoTModel Scale=Medium, Framework=LoT, Visual Highlighting=true2026.04 | 84.3 | — | |
| InternVL3.5-8BModel Scale=Medium, Framework=Baseline, Visual Highlighting=false2026.04 | 84.3 | — | |
| Qwen2.5-VL-7BRetained Tokens=1225, Token Retention Ratio=100%2026.03 | 84.1 | 100 | |
| InternVL3.5Size=8B2026.03 | 84 | — | |
| InternVL3.5Size=8B, mode=instruct2026.04 | 84 | — | |
| Q-MaskSize=2B2026.03 | 83.3 | — | |
| V-SkipBase Model=Qwen2.5-VL-7B, Hyperparameter (K/N)=N=72026.06 | 83.1 | — | |
| InternVL3.5-8B + LoTModel Scale=Medium, Framework=LoT, Visual Highlighting=true2026.04 | 83 | — | |
| InternVL3.5-4BModel Scale=Small, Framework=Baseline, Visual Highlighting=false2026.04 | 82.9 | — | |
| InternVL3.5-4B + LoTModel Scale=Small, Framework=LoT, Visual Highlighting=true2026.04 | 82.5 | — | |
| GLM-4.1VSize=9B2026.03 | 82.3 | — | |
| InternVL3.5Size=4B2026.03 | 82.2 | — | |
| TokenVL-2BSize=2B2026.03 | 82.1 | — | |
| MarternSize=8B2026.03 | 82 | — | |
| Qwen2-VL-7B + LoTModel Scale=Medium, Framework=LoT, Visual Highlighting=true2026.04 | 81.3 | — | |
| GPT 5.2Size=Closed2026.03 | 80.7 | — | |
| Granite-Vision -2B2025.10 | 80 | — | |
| Granite-Vision -2B + CARESCost=-68%2025.10 | 80 | — | |
| Qwen2-VL-7BModel Scale=Medium, Framework=Baseline, Visual Highlighting=false2026.04 | 80 | — | |
| ShortVBase Model=Qwen2.5-VL-7B, Hyperparameter (K/N)=N=72026.06 | 80 | — | |
| Qwen2.5-VLSize=3B2026.03 | 79.7 | — | |
| Qwen2.5-VL-3B + LoTModel Scale=Small, Framework=LoT, Visual Highlighting=true2026.04 | 79.7 | — | |
| Qwen3-VL-8BModel Scale=Medium, Framework=Baseline, Visual Highlighting=false2026.04 | 79.5 | — | |
| Qwen2.5-VL-32B + LoTModel Scale=Large, Framework=LoT, Visual Highlighting=true2026.04 | 79 | — | |
| Claude-3.5-SonnetSize=Closed2026.03 | 78.8 | — | |
| Qwen2.5-VL-32BModel Scale=Large, Framework=Baseline, Visual Highlighting=false2026.04 | 78.7 | — | |
| DivPruneRetained Tokens=512, Token Retention Ratio=60.5% Reduction2026.03 | 78.5 | 96.8 | |
| AgilePrunerRetained Tokens=512, Token Retention Ratio=60.5% Reduction2026.03 | 78.5 | 96.9 | |
| Qwen3-VL-4B + LoTModel Scale=Small, Framework=LoT, Visual Highlighting=true2026.04 | 77.4 | — | |
| Granite-Vision -2B + CARES-ARCost=-75%2025.10 | 77 | — | |
| GPT-4o2025.10 | 77 | — | |
| Qwen2.5-VL-3BModel Scale=Small, Framework=Baseline, Visual Highlighting=false2026.04 | 76.1 | — | |
| Qwen2.5-VL-72B + CARESCost=-64%2025.10 | 76 | — | |
| Qwen3-VL-4BModel Scale=Small, Framework=Baseline, Visual Highlighting=false2026.04 | 76 | — | |
| Qwen2.5-VL-72B2025.10 | 75 | — | |
| GPT-4o + CARESCost=-33%2025.10 | 75 | — | |
| Qwen2-VL-2BModel Scale=Small, Framework=Baseline, Visual Highlighting=false2026.04 | 74.6 | — | |
| VTWBase Model=Qwen2.5-VL-7B, Hyperparameter (K/N)=K=72026.06 | 74.6 | — | |
| Qwen2-VL-2B + LoTModel Scale=Small, Framework=LoT, Visual Highlighting=true2026.04 | 74 | — | |
| GPT-4oSize=Closed2026.03 | 73.6 | — | |
| Gemma3Size=27B2026.03 | 71.7 | — | |
| Claude-3-OpusSize=Closed2026.03 | 69.4 | — | |
| DivPruneRetained Tokens=256, Token Retention Ratio=79.1% Reduction2026.03 | 66.4 | 90.7 | |
| AgilePrunerRetained Tokens=256, Token Retention Ratio=79.1% Reduction2026.03 | 65.8 | 92.1 | |
| DocOwl-1.5-ChatSize=8B2026.03 | 59.9 | — | |
| TextMonkeySize=8B2026.03 | 56.1 | — | |
| LLaVA-NeXT-13BBase Model=LLaVA-NeXT-13B2026.06 | 55.1 | — | |
| V-SkipBase Model=LLaVA-NeXT-13B, Hyperparameter (K/N)=N=242026.06 | 53.9 | — | |
| LLaVA-NeXT-7BBase Model=LLaVA-NeXT-7B2026.06 | 52.3 | — | |
| ShortVBase Model=LLaVA-NeXT-13B, Hyperparameter (K/N)=N=242026.06 | 49.6 | — | |
| V-SkipBase Model=LLaVA-NeXT-7B, Hyperparameter (K/N)=N=192026.06 | 49.4 | — | |
| ShortVBase Model=LLaVA-NeXT-7B, Hyperparameter (K/N)=N=192026.06 | 45.6 | — | |
| VTWBase Model=LLaVA-NeXT-13B, Hyperparameter (K/N)=K=202026.06 | 35.5 | — | |
| LLaVA-1.5-13BBase Model=LLaVA-1.5-13B2026.06 | 33.7 | — | |
| V-SkipBase Model=LLaVA-1.5-13B, Hyperparameter (K/N)=N=242026.06 | 33 | — | |
| ShortVBase Model=LLaVA-1.5-13B, Hyperparameter (K/N)=N=242026.06 | 32.1 | — | |
| V-SkipBase Model=LLaVA-1.5-7B, Hyperparameter (K/N)=N=192026.06 | 31.6 | — | |
| LLaVA-1.5-7BBase Model=LLaVA-1.5-7B2026.06 | 31.3 | — | |
| ShortVBase Model=LLaVA-1.5-7B, Hyperparameter (K/N)=N=192026.06 | 29.5 | — | |
| EOSRetain (visual tokens)=322026.06 | 25.5 | — | |
| SparseVLM-v2Retain (visual tokens)=322026.06 | 24.9 | — | |
| VTWBase Model=LLaVA-1.5-13B, Hyperparameter (K/N)=K=202026.06 | 24.2 | — | |
| SparseVLM-v1Retain (visual tokens)=322026.06 | 24.1 | — | |
| ToMeRetain (visual tokens)=322026.06 | 21.1 | — | |
| EOSRetain (visual tokens)=162026.06 | 20.8 | — | |
| SparseVLM-v2Retain (visual tokens)=162026.06 | 20.3 | — | |
| SparseVLM-v1Retain (visual tokens)=162026.06 | 19.4 | — | |
| ToMeRetain (visual tokens)=162026.06 | 17.4 | — | |
| VTWBase Model=LLaVA-NeXT-7B, Hyperparameter (K/N)=K=162026.06 | 6.5 | — | |
| VTWBase Model=LLaVA-1.5-7B, Hyperparameter (K/N)=K=162026.06 | 5.1 | — | |
| PoolRetain (visual tokens)=322026.06 | 2.1 | — | |
| PoolRetain (visual tokens)=162026.06 | 1.5 | — |