OCR Evaluation on OCRBench
909ScoreGemini 3-Pro
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini 3-Pro2026.02 | 909 | — | |
| SAIL-VL2Number of Parameters=2.7B2025.12 | 895 | — | |
| Qwen2.5-VL-72BModel Size=72B2025.02 | 885 | — | |
| Qwen 2.5-VL 72Bthinking=false, decoding=greedy2025.05 | 885 | — | |
| SAIL-VL1.5Number of Parameters=2.5B2025.12 | 885 | — | |
| Qwen-VL-Max-0809Params (B)=722024.10 | 881 | — | |
| Seed 1.5-VLthinking=false, decoding=greedy2025.05 | 881 | — | |
| InternVL2.5-MPOModel Size=8B2025.01 | 880 | — | |
| 1.7B K^8_{0.5}Size=18B-A3B, Sparsity Configuration=K=8, alpha=0.52026.01 | 880 | — | |
| ERNIE 5.02026.02 | 878 | — | |
| ERNIE 5.0-BaseModel type=pre-trained2026.02 | 875 | — | |
| 1.7B K^4_{1.0}Size=18B-A3B, Sparsity Configuration=K=4, alpha=1.02026.01 | 873 | — | |
| InternVL3.5-20B-A4BSize=20B-A4B2026.01 | 870 | — | |
| Ovis2Number of Parameters=2.5B2025.12 | 868 | — | |
| Qwen2-VLSize=7B, #token /image tile=Native resolution (unfixed)2024.12 | 866 | — | |
| Gemini 1.5 Prothinking=true, decoding=greedy2025.05 | 866 | — | |
| Gemini 2.5-Pro2026.02 | 866 | — | |
| Qwen2.5-VL-7BModel Size=7B2025.02 | 864 | — | |
| Qwen3-VLmode=Thinking2026.02 | 863 | — | |
| Seed 1.5-VLthinking=true, decoding=greedy2025.05 | 861 | — | |
| HyperVL ViTLNumber of Parameters=2.0B2025.12 | 859 | — | |
| Qwen3-VLNumber of Parameters=2.1B2025.12 | 858 | — | |
| QADModel=Llama Nemotron Nano 12B v2 VL2026.01 | 858 | — | |
| Isaac 0.2Size=2B2026.01 | 857 | — | |
| BaselineModel=Llama Nemotron Nano 12B v2 VL2026.01 | 855 | — | |
| InternVL2.5-78BModel Size=78B2025.02 | 854 | — | |
| PTQModel=Llama Nemotron Nano 12B v2 VL2026.01 | 850 | — | |
| QATModel=Llama Nemotron Nano 12B v2 VL2026.01 | 848 | — | |
| InternVL2.5-MPOModel Size=2B2025.01 | 846 | — | |
| AndesVLNumber of Parameters=2.4B2025.12 | 846 | — | |
| EMOVAModel Size=72B2024.09 | 843 | — | |
| InternVL2-Llama3-76BParams (B)=762024.10 | 842 | — | |
| InternVL3.5Number of Parameters=2.3B2025.12 | 836 | — | |
| SAIL-VLModel Size=8B2025.01 | 835 | — | |
| InternVL3Number of Parameters=2.1B2025.12 | 835 | — | |
| DeepSeekVL-2Model Size=8B2025.01 | 834 | — | |
| Qwen2-VLModel Size=8B2025.01 | 833 | — | |
| HyperVLNumber of Parameters=1.8B2025.12 | 830 | — | |
| EMOVAModel Size=7B2024.09 | 814 | — | |
| Qwen2-VLSize=2B, #token /image tile=Native resolution (unfixed)2024.12 | 809 | — | |
| Qwen2-VLNumber of Parameters=2.2B2025.12 | 809 | — | |
| DeepSeekVL-2Model Size=2B2025.01 | 808 | — | |
| PVC InternVL2 (Ours)Size=8B, #token /image tile=642024.12 | 807 | — | |
| GPT-4othinking=false, decoding=greedy2025.05 | 806 | — | |
| SAIL-VLModel Size=2B2025.01 | 806 | — | |
| GPT-4o-202408062024.10 | 805 | — | |
| GPT-5tier=High2026.02 | 804 | — | |
| EMOVAModel Size=3B2024.09 | 803 | — | |
| Qwen2-VL-2BParams (B)=2.12024.10 | 797 | — | |
| Qwen2.5-VL-3BModel Size=3B2025.02 | 797 | — | |
| PVC InternVL2 (Ours)Size=2B, #token /image tile=642024.12 | 796 | — | |
| MiniMonkeyParams (B)=2.22024.10 | 794 | — | |
| InternVL2Size=8B, #token /image tile=2562024.12 | 794 | — | |
| Qwen2-VL-2B#Params=2.2B2024.12 | 794 | — | |
| Qwen2-VLModel Size=2B2025.01 | 794 | — | |
| Claude 3.7 Sonnetthinking=true, decoding=sampling2025.05 | 793 | — | |
| Claude3.5-Sonnet2024.10 | 788 | — | |
| Claude-3.5 Sonnet2025.02 | 788 | — | |
| InternVL2Size=2B, #token /image tile=2562024.12 | 784 | — | |
| InternVL2#Params=2.2B2024.12 | 784 | — | |
| H2OVL-Mississippi-2BParams (B)=2.12024.10 | 782 | — | |
| InternVL2-2BParams (B)=2.12024.10 | 781 | — | |
| InternVL2-1BParams (B)=0.82024.10 | 755 | — | |
| Gemini-1.5-Pro2024.10 | 754 | — | |
| Gemini 1.5 Pro2025.02 | 754 | — | |
| VITAModel Size=1.52024.09 | 752 | — | |
| H2OVL-Mississippi-0.8BParams (B)=0.82024.10 | 751 | — | |
| OpenAI o1thinking=true, decoding=greedy2025.05 | 750 | — | |
| Oryx-1.5Size=32B2024.09 | 746 | — | |
| OryxSize=34B2024.09 | 743 | — | |
| HoVLE#Params=2.6B, Resolution=HD2024.12 | 740 | — | |
| GPT-4o2024.12 | 736 | — | |
| GPT-4o2025.02 | 736 | — | |
| GPT-4o2024.09 | 736 | — | |
| HoVLE#Params=2.6B, Resolution=Standard2024.12 | 735 | — | |
| Qwen-VL-Plus2024.10 | 726 | — | |
| MiniCPM-Llama3-V 2.5Size=8.5B2024.08 | 725 | — | |
| InternVL 1.5#param=26B, open-source=false2024.04 | 724 | — | |
| Qwen-VL-Maxopen-source=false2024.04 | 723 | — | |
| Oryx-1.5Size=7B2024.09 | 713 | — | |
| Baichuan-OmniModel Size=7B2024.09 | 700 | — | |
| Qwen-VL-Plusopen-source=false2024.04 | 694 | — | |
| Claude-3 Opusopen-source=false2024.04 | 694 | — | |
| IXC-2.5Size=7B, #token /image tile=4002024.12 | 690 | — | |
| Emu3-ChatLLM=8B from scratch, Category=Understanding Only2024.12 | 687 | — | |
| Emu3#Params=8B2024.12 | 687 | — | |
| Gemini ProProprietary=true2024.08 | 680 | — | |
| Gemini-1.0-Pro2024.10 | 680 | — | |
| GPT-4v (1106, detail-high)2024.10 | 678 | — | |
| VITASize=8x7B2024.09 | 678 | — | |
| VITAModel Size=8x7B2024.09 | 678 | — | |
| Oryx MLLMSize=7B2024.12 | 672 | — | |
| OryxSize=7B2024.09 | 672 | — | |
| ILLUMELLM=Vicuna-7B, Category=Unify Understanding and Generation2024.12 | 669 | — | |
| Gemini ProLLM Size=Unk2024.03 | 659 | — | |
| GPT-4VLLM Size=Unk2024.03 | 659 | — | |
| Gemini Pro 1.0open-source=false2024.04 | 659 | — | |
| Claude-3 Haikuopen-source=false2024.04 | 658 | — | |
| GPT-4V2024.09 | 656 | — | |
| InternVL1.5#Params=2.2B2024.12 | 654 | — |