OCR & Document Understanding on OCRBench
911ScoreOpenVLThinkerV2
Evaluation Results
| Method | Links | |
|---|---|---|
| OpenVLThinkerV22026.04 | 911 | |
| Qwen3-VL GDPOAlignment=GDPO2026.04 | 897 | |
| VisionZero2026.04 | 885 | |
| DeepEyesV22026.04 | 882 | |
| Qwen3-VL GRPOAlignment=GRPO2026.04 | 875 | |
| Gemini 2.5 Pro2026.04 | 866 | |
| OneThinker-8BParameters=8B2026.04 | 833 | |
| InstructVLA-Generalistnumber of parameters=1.5B, robot state=true2025.07 | 829 | |
| Qwen3-VL-InstructModel Type=Instruct2026.04 | 819 | |
| Eagle2number of parameters=1.5B2025.07 | 818 | |
| InstructVLA-Generalistnumber of parameters=1.5B2025.07 | 814 | |
| GPT-52026.04 | 810 | |
| Qwen2-VLnumber of parameters=1.5B2025.07 | 809 | |
| VisionThink2026.04 | 808 | |
| ChatVLAnumber of parameters=1.5B2025.07 | 729 | |
| PaliGemmanumber of parameters=2B2025.07 | 614 | |
| Magmanumber of parameters=8B2025.07 | 518 | |
| Bunnynumber of parameters=8B2025.07 | 444 | |
| MIRROR (ours)Param Size=7B2026.02 | 92 | |
| Qwen3.5-OmniOpen-Source=false, Size=Flash2026.04 | 89.1 | |
| MIRROR (w/o tool)Param Size=7B2026.02 | 88.3 | |
| Nemotron 3 Nano OmniOpen-Source=true, Size=30B-A3B, Reasoning=off2026.04 | 88.3 | |
| InternVL3.5-20B-A4BModel=InternVL3.5-20B-A4B2026.05 | 86.7 | |
| Nemotron 3 Nano OmniOpen-Source=true, Size=30B-A3B, Reasoning=on2026.04 | 86.6 | |
| Qwen2.5-VLScale=7B2026.06 | 86.4 | |
| Qwen3-OmniOpen-Source=true, Size=30B-A3B2026.04 | 86 | |
| Nemotron Nano V2 VLOpen-Source=true, Size=12B, Reasoning=off2026.04 | 85.6 | |
| Qwen3.5-4BModel=Qwen3.5-4B2026.05 | 85.3 | |
| InternVL3Param Size=8B2026.02 | 85 | |
| Qwen3-VL-2BModel=Qwen3-VL-2B2026.05 | 84.1 | |
| Qwen3-VL-4BModel=Qwen3-VL-4B2026.05 | 84.1 | |
| VanillaBackbone=Qwen2.5-VL-7B, TFLOPs=100%2026.06 | 84.1 | |
| Nemotron Nano V2 VLOpen-Source=true, Size=12B, Reasoning=on2026.04 | 83.5 | |
| InternVL3.5-2BModel=InternVL3.5-2B2026.05 | 83.4 | |
| DeepSeek-VL2-16B-A2.4BModel=DeepSeek-VL2-16B-A2.4B2026.05 | 83.3 | |
| Qwen2.5-VL-7BParam Size=7B2026.02 | 83.2 | |
| Qwen3.5-2BModel=Qwen3.5-2B2026.05 | 83.1 | |
| LLaVA-OVScale=1.5-8B2026.06 | 82.9 | |
| VanillaBackbone=Qwen3-VL-8B, TFLOPs=100%2026.06 | 82.8 | |
| Qwen2.5-VL-3BParam Size=3B2026.02 | 82.6 | |
| Qwen2.5-VL-3BModel=Qwen2.5-VL-3B2026.05 | 82.5 | |
| InternVL3Param Size=2B2026.02 | 82.2 | |
| InternVL3.5-4BModel=InternVL3.5-4B2026.05 | 82 | |
| PLM-3BModel=PLM-3B2026.05 | 80.6 | |
| Op-SkipBackbone=Qwen3-VL-8B, TFLOPs=66%, N=202026.06 | 80.4 | |
| ZAYA1-VL-8B-A1BModel=ZAYA1-VL-8B-A1B2026.05 | 79.8 | |
| Qwen2.5-Omni-3BModel Scale=3B2026.06 | 78.2 | |
| Op-SkipBackbone=Qwen2.5-VL-7B, TFLOPs=55%, N=202026.06 | 76.9 | |
| Llama-3.2 11B-VisionCheckpoint=base2026.06 | 75.9 | |
| AVLM-2BModel Scale=2B2026.06 | 75.9 | |
| APETBackbone=Qwen3-VL-8B, TFLOPs=66%2026.06 | 75.8 | |
| ShortVBackbone=Qwen3-VL-8B, TFLOPs=58%, N=202026.06 | 73.9 | |
| Gemma-4-E2B-itModel Scale=4B2026.06 | 72.8 | |
| VSkipBackbone=Qwen3-VL-8B, TFLOPs=76%, N=202026.06 | 72.2 | |
| MOSS-Video-PreviewSFT Mode=offline2026.06 | 70.5 | |
| VSkipBackbone=Qwen2.5-VL-7B, TFLOPs=67%, N=162026.06 | 69.4 | |
| MOSS-Video-PreviewSFT Mode=real-time2026.06 | 67.7 | |
| V2DropBackbone=Qwen3-VL-8B, TFLOPs=66%2026.06 | 63.8 | |
| LLaVA-OneVisionParam Size=7B2026.02 | 62.1 | |
| Molmo2-4BModel=Molmo2-4B2026.05 | 62 | |
| APETBackbone=Qwen2.5-VL-7B, TFLOPs=55%2026.06 | 60 | |
| V2DropBackbone=Qwen2.5-VL-7B, TFLOPs=55%2026.06 | 56.4 | |
| MolmoE-8B-A1BModel=MolmoE-8B-A1B2026.05 | 55 | |
| ShortVBackbone=Qwen2.5-VL-7B, TFLOPs=50%, N=202026.06 | 45.5 | |
| VTWBackbone=Qwen2.5-VL-7B, TFLOPs=50%, K=162026.06 | 42.39 | |
| VTWBackbone=Qwen3-VL-8B, TFLOPs=58%, K=162026.06 | 37 | |
| VanillaBackbone=LLaVA-1.5-7B, TFLOPs=100%2026.06 | 31.5 | |
| VSkipBackbone=LLaVA-1.5-7B, TFLOPs=76%, N=202026.06 | 30.8 | |
| VSkip+Backbone=LLaVA-1.5-7B, TFLOPs=76%, N=202026.06 | 30.5 | |
| Op-SkipBackbone=LLaVA-1.5-7B, TFLOPs=66%, N=202026.06 | 30.2 | |
| APETBackbone=LLaVA-1.5-7B, TFLOPs=66%2026.06 | 29.9 | |
| V2DropBackbone=LLaVA-1.5-7B, TFLOPs=66%2026.06 | 29.7 | |
| ShortVBackbone=LLaVA-1.5-7B, TFLOPs=55%, N=202026.06 | 28.1 | |
| VTWBackbone=LLaVA-1.5-7B, TFLOPs=55%, K=162026.06 | 5.2 | |
| OpenVLAnumber of parameters=7B, fine-tuned=true2025.07 | 2.5 | |
| OpenVLAnumber of parameters=7B2025.07 | 0 | |
| ECoTnumber of parameters=7B2025.07 | 0 |