Document Visual Question Answering on MMLongbench doc
45.6AccuracyGPT-4.1
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GPT-4.1Size=-, Paradigm=End-to-End2025.12 | 45.6 | 49.7 | — | — | — | — | — | — | — | — | — | |
| OracleTop-k retrieval=3, Input modality=I + T, Judge model=Qwen2.5-7B-Instruct, Backbone Vision Language Model=Qwen2.5-VL-7B-Instruct2025.09 | 43.25 | — | — | — | — | — | — | — | — | — | — | |
| GPT-4oSize=-, Paradigm=End-to-End2025.12 | 41.6 | 42.3 | — | — | — | — | — | — | — | — | — | |
| OracleTop-k retrieval=3, Input modality=I, Judge model=Qwen2.5-7B-Instruct, Backbone Vision Language Model=Qwen2.5-VL-7B-Instruct2025.09 | 41.13 | — | — | — | — | — | — | — | — | — | — | |
| Lang2ActCategory=Tool-Enhanced VLMs2026.01 | 36.55 | — | 38.59 | 31.8 | 31.46 | 32.23 | 33.05 | — | — | — | — | |
| Claude-3.7-SonnetSize=-, Paradigm=End-to-End2025.12 | 33.9 | 38.4 | — | — | — | — | — | — | — | — | — | |
| Pixel-ReasonerCategory=Tool-Enhanced VLMs2026.01 | 33.22 | — | 37.37 | 27.19 | 24.72 | 32.89 | 27.97 | — | — | — | — | |
| CogDocSize=7B, Paradigm=Unified Thinking2025.12 | 33 | 31.5 | — | — | — | — | — | — | — | — | — | |
| OpenVLThinkerCategory=Vision-Language Reasoning Models (VLRMs)2026.01 | 32.71 | — | 35.23 | 28.57 | 26.97 | 30.23 | 27.97 | — | — | — | — | |
| Vision-R1Category=Vision-Language Reasoning Models (VLRMs)2026.01 | 32.25 | — | 32.55 | 27.65 | 26.4 | 29.9 | 30.51 | — | — | — | — | |
| DOCPRUNEBackbone=Qwen2.5-VL (7B), # Pages=42026.04 | 32.2 | 33.7 | — | — | — | — | — | 22 | 32.5 | 2 | 1.6 | |
| ThinkLite-VLCategory=Vision-Language Reasoning Models (VLRMs)2026.01 | 31.66 | — | 33.22 | 24.88 | 26.97 | 29.9 | 27.12 | — | — | — | — | |
| VisionMattersCategory=Vision-Language Reasoning Models (VLRMs)2026.01 | 31.66 | — | 35.91 | 23.96 | 29.21 | 27.91 | 28.81 | — | — | — | — | |
| VRAG-RLCategory=Tool-Enhanced VLMs2026.01 | 31.55 | — | 31.54 | 24.88 | 30.34 | 30.23 | 20.34 | — | — | — | — | |
| Qwen2.5-VL (7B)Backbone=Qwen2.5-VL (7B), # Pages=42026.04 | 31.5 | 33.4 | — | — | — | — | — | 55 | 80.5 | 1 | 0.7 | |
| SigLIPTop-k retrieval=3, Judge model=Qwen2.5-7B-Instruct, Backbone Vision Language Model=Qwen2.5-VL-7B-Instruct2025.09 | 31.33 | — | — | — | — | — | — | — | — | — | — | |
| FRAGSize=8B, Paradigm=RAG2025.12 | 31.3 | 30.7 | — | — | — | — | — | — | — | — | — | |
| CMRAGTop-k retrieval=3, Judge model=Qwen2.5-7B-Instruct, Backbone Vision Language Model=Qwen2.5-VL-7B-Instruct2025.09 | 31.05 | — | — | — | — | — | — | — | — | — | — | |
| EVisRAGCategory=Multimodal Retrieval-Augmented Generation Models (MRAGs)2026.01 | 30.97 | — | 32.55 | 27.19 | 25.28 | 27.57 | 27.12 | — | — | — | — | |
| CLIP-L/14-336Top-k retrieval=3, Judge model=Qwen2.5-7B-Instruct, Backbone Vision Language Model=Qwen2.5-VL-7B-Instruct2025.09 | 30.68 | — | — | — | — | — | — | — | — | — | — | |
| R1-OnevisionCategory=Vision-Language Reasoning Models (VLRMs)2026.01 | 30.5 | — | 30.54 | 26.73 | 27.53 | 27.57 | 25.42 | — | — | — | — | |
| MM-Search-R1Category=Multimodal Retrieval-Augmented Generation Models (MRAGs)2026.01 | 29.92 | — | 31.54 | 25.35 | 27.53 | 26.91 | 22.88 | — | — | — | — | |
| InternVL3-8BSize=8B, Paradigm=End-to-End2025.12 | 29.6 | 28.8 | — | — | — | — | — | — | — | — | — | |
| DocopilotSize=8B, Paradigm=End-to-End2025.12 | 28.8 | 23 | — | — | — | — | — | — | — | — | — | |
| GPT-4o miniSize=-, Paradigm=End-to-End2025.12 | 28.6 | 29.4 | — | — | — | — | — | — | — | — | — | |
| DOCPRUNEBackbone=Qwen2-VL (7B), # Pages=42026.04 | 28.5 | 30.5 | — | — | — | — | — | 22 | 28.6 | 1.2 | 1.8 | |
| DOCPRUNEBackbone=Qwen2.5-VL (7B), # Pages=12026.04 | 28.2 | 28.3 | — | — | — | — | — | 5.2 | 7.2 | 8.2 | 6.1 | |
| TOTCategory=Prompting Methods2026.01 | 27.82 | — | 27.18 | 22.12 | 21.91 | 27.57 | 25.42 | — | — | — | — | |
| DirectCategory=Prompting Methods2026.01 | 27.59 | — | 31.21 | 23.5 | 27.53 | 23.59 | 21.19 | — | — | — | — | |
| MiMo-VL-7B-SFTSize=7B, Paradigm=End-to-End2025.12 | 27.5 | 27.2 | — | — | — | — | — | — | — | — | — | |
| Qwen2-VL (7B)Backbone=Qwen2-VL (7B), # Pages=42026.04 | 27.4 | 29.3 | — | — | — | — | — | 55 | 80.5 | 0.6 | 0.7 | |
| Qwen2.5-VL (7B)Backbone=Qwen2.5-VL (7B), # Pages=12026.04 | 27.3 | 27.9 | — | — | — | — | — | 13.7 | 16.9 | 3.9 | 2.9 | |
| GOTCategory=Prompting Methods2026.01 | 27.12 | — | 29.19 | 21.66 | 22.47 | 24.25 | 22.03 | — | — | — | — | |
| VisDomCategory=Multimodal Retrieval-Augmented Generation Models (MRAGs)2026.01 | 25.96 | — | 33.89 | 21.2 | 21.35 | 18.94 | 21.19 | — | — | — | — | |
| VTWBackbone=Qwen2-VL (7B), # Pages=42026.04 | 25.3 | 26.8 | — | — | — | — | — | 55 | 35.8 | 0.6 | 1.4 | |
| DOCPRUNEBackbone=Qwen2-VL (7B), # Pages=12026.04 | 25 | 26.8 | — | — | — | — | — | 5.2 | 7.1 | 5.7 | 6.1 | |
| Qwen2.5-VL-72B-InstructSize=72B, Paradigm=End-to-End2025.12 | 24.9 | 24.6 | — | — | — | — | — | — | — | — | — | |
| Qwen2-VL (7B)Backbone=Qwen2-VL (7B), # Pages=12026.04 | 24.5 | 25.8 | — | — | — | — | — | 13.7 | 16.9 | 2.2 | 2.9 | |
| DivPruneBackbone=Qwen2-VL (7B), # Pages=12026.04 | 24.3 | 24.3 | — | — | — | — | — | 13.7 | 15.1 | 2.5 | 3.2 | |
| VTWBackbone=Qwen2-VL (7B), # Pages=12026.04 | 23.2 | 24 | — | — | — | — | — | 13.7 | 8.3 | 2.5 | 5.4 | |
| SV-RAGSize=8B, Paradigm=RAG2025.12 | 23 | 24.2 | — | — | — | — | — | — | — | — | — | |
| DivPruneBackbone=Qwen2-VL (7B), # Pages=42026.04 | 23 | 23.2 | — | — | — | — | — | 55 | 66.2 | 0.6 | 0.8 | |
| FastVBackbone=Qwen2-VL (7B), # Pages=42026.04 | 22.8 | 24.4 | — | — | — | — | — | 55 | 40.6 | 0.6 | 1.3 | |
| Qwen2.5-VL-7B-InstructSize=7B, Paradigm=End-to-End2025.12 | 21.2 | 20.7 | — | — | — | — | — | — | — | — | — | |
| M3DocRAGSize=10B, Paradigm=RAG2025.12 | 21 | 22.6 | — | — | — | — | — | — | — | — | — | |
| FastVBackbone=Qwen2-VL (7B), # Pages=12026.04 | 20.4 | 21.1 | — | — | — | — | — | 13.7 | 10.1 | 2.5 | 4.5 | |
| VisRAGSize=8B, Paradigm=RAG2025.12 | 18.8 | 18.3 | — | — | — | — | — | — | — | — | — | |
| mPLUG-DocOwl2Size=8B, Paradigm=End-to-End2025.12 | 13.4 | 8.9 | — | — | — | — | — | — | — | — | — |