Multimodal Document Question Answering on MMLongBench-Doc
65.8Overall AccuracyHuman
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Human2026.04 | 65.8 | — | — | — | — | — | — | — | — | 66 | |
| Human Baseline2026.06 | 65.8 | — | — | — | — | — | — | — | — | 66 | |
| DocAgentModel=Claude 3.5 Sonnet2026.06 | 57.3 | — | — | — | — | — | — | — | — | 54.1 | |
| MARDocModel=Qwen3-VL-30B-A3B-Instruct2026.06 | 57.1 | — | — | — | — | — | — | — | — | 54.6 | |
| DocDancerModel=Qwen3-VL-30B-A3B-Thinking(ft)2026.06 | 54.4 | — | — | — | — | — | — | — | — | 53.9 | |
| SLEUTHModel=Qwen3-VL-8B-Instruct2026.06 | 52.8 | — | — | — | — | — | — | — | — | — | |
| MARDocModel=Qwen3-VL-8B-Instruct2026.06 | 52.7 | — | — | — | — | — | — | — | — | 50.3 | |
| DocDancerModel=GPT-4o2026.06 | 52.3 | — | — | — | — | — | — | — | — | 50.8 | |
| DocAgentModel=GPT-4o2026.06 | 51.8 | — | — | — | — | — | — | — | — | 49.1 | |
| MM-Doc-R1Backbone=Qwen3-8B, Reinforcement Learning Algorithm=SPO2026.04 | 49.7 | 46.2 | 38.1 | 40.8 | 52.8 | 35.9 | 56 | 31.2 | 68.2 | 46.1 | |
| Qwen3-8BBackbone=Qwen3-8B, Ground-Truth Evidence=true2026.04 | 49.6 | 44.3 | 37.7 | 25.7 | 59.3 | 22.8 | 42.7 | 35.7 | 89.2 | 46.9 | |
| DocAgentModel=Qwen3-VL-30B-A3B-Instruct2026.06 | 48.3 | — | — | — | — | — | — | — | — | 45.8 | |
| MLDocRAGRetriever=BGE-m3, Generator=LVLM, Modality (TXT)=true, Modality (IMG)=true2026.02 | 47.9 | 47.2 | 37.8 | 42.7 | 41.3 | 31.9 | 52.6 | 26.4 | 71.5 | — | |
| MACTModel=MiMo-VL-7B2026.06 | 47.4 | — | — | — | — | — | — | — | — | — | |
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B, Ground-Truth Evidence=true2026.04 | 46.8 | 33.8 | 38.7 | 31.8 | 32.3 | 34.1 | 46.6 | 20.5 | 92.8 | 41.7 | |
| DocAgentModel=Qwen3-VL-8B-Instruct2026.06 | 46.6 | — | — | — | — | — | — | — | — | 42.6 | |
| RAG^txt+txt2i_BGE-m3Retriever=BGE-m3, Generator=LLM, Modality (TXT)=true, Modality (IMG)=false2026.02 | 46.5 | 42.3 | 26.1 | 37.6 | 45.41 | 31.3 | 49 | 27.2 | 71.5 | — | |
| LC^txt+imgRetriever=-, Generator=LVLM, Modality (TXT)=true, Modality (IMG)=true2026.02 | 46.1 | 53.8 | 34.5 | 45.5 | 45.9 | 35.9 | 52.4 | 37.8 | 45.6 | — | |
| MM-Doc-R1Backbone=Qwen3-4B, Reinforcement Learning Algorithm=SPO2026.04 | 46 | 41.1 | 37.2 | 35.6 | 47.2 | 30.5 | 50.5 | 27.5 | 68 | 41.2 | |
| Vanilla VLMModel=Qwen3-VL-30B-A3B-Instruct2026.06 | 45.9 | — | — | — | — | — | — | — | — | 45.6 | |
| MM-Doc-R1Backbone=Qwen3-8B, Reinforcement Learning Algorithm=None2026.04 | 45.7 | 39.6 | 37.6 | 37.8 | 45.6 | 27.3 | 47.3 | 27.5 | 73.9 | 41.5 | |
| Qwen3-VL-32B + CDSParams=32B, Retriever=CDS2026.06 | 45.01 | — | — | — | — | — | — | — | — | — | |
| MM-Doc-R1Backbone=Qwen3-8B, Reinforcement Learning Algorithm=GRPO2026.04 | 44.7 | 40.9 | 36.8 | 35.7 | 49.9 | 28.3 | 48.5 | 30.2 | 60.5 | 41.9 | |
| BOOKRAGModel=Qwen2.5-VL-30B2026.06 | 43.8 | — | — | — | — | — | — | — | — | 44.9 | |
| RAG^txt+txt2i_BM25Retriever=BM25, Generator=LLM, Modality (TXT)=true, Modality (IMG)=false2026.02 | 43.7 | 43.6 | 31.9 | 38.2 | 38.1 | 27 | 45.3 | 26.9 | 66.7 | — | |
| MACTModel=Qwen2.5-VL-7B2026.06 | 43.7 | — | — | — | — | — | — | — | — | — | |
| Vanilla VLMModel=GPT-4o2026.06 | 42.8 | — | — | — | — | — | — | — | — | 44.9 | |
| RAGAnythingModel=GPT-4o-mini2026.06 | 42.8 | — | — | — | — | — | — | — | — | — | |
| GraphRAG^txt+txt2i_BGE-m3Retriever=BGE-m3, Generator=LLM, Modality (TXT)=true, Modality (IMG)=false2026.02 | 42.6 | 48.1 | 25.5 | 36.4 | 51.5 | 30.2 | 51.3 | 27.1 | 48.5 | — | |
| LC^txt+txt2iRetriever=-, Generator=LLM, Modality (TXT)=true, Modality (IMG)=false2026.02 | 42.5 | 48.5 | 32.8 | 40.5 | 41.7 | 28.6 | 44.1 | 32.5 | 54.8 | — | |
| MDocAgentModel=GPT-4o2026.06 | 42 | — | — | — | — | — | — | — | — | — | |
| Vanilla VLMModel=Qwen3-VL-8B-Instruct2026.06 | 41.9 | — | — | — | — | — | — | — | — | 42.2 | |
| VRAG^page_ColPaliRetriever=ColPali, Generator=LVLM, Modality (TXT)=false, Modality (IMG)=true2026.02 | 41.4 | 40 | 37.8 | 32 | 30.7 | 32.2 | 44.1 | 23.9 | 63.2 | — | |
| GLM-4.1VParams=9B, Retriever=None (Single-VLM)2026.06 | 41.04 | — | — | — | — | — | — | — | — | — | |
| MoLoRAGModel=Qwen2.5-VL-7B2026.06 | 41 | — | — | — | — | — | — | — | — | — | |
| Lang2ActCategory=Tool-Enhanced VLMs2026.01 | 40.51 | 43.96 | 38.14 | 36.52 | 27.19 | 40.86 | — | — | — | — | |
| Vision-R1Category=Vision-Language Reasoning Models (VLRMs)2026.01 | 40.28 | 41.28 | 42.37 | 38.2 | 28.11 | 40.2 | — | — | — | — | |
| Qwen3-VL-32BParams=32B, Retriever=None (Single-VLM)2026.06 | 40.19 | — | — | — | — | — | — | — | — | — | |
| DocSeekerModel=Qwen-2.5-VL-7B-Instruct2026.06 | 40.1 | — | — | — | — | — | — | — | — | 38.4 | |
| MM-Doc-R1Backbone=Qwen3-4B, Reinforcement Learning Algorithm=GRPO2026.04 | 39.9 | 36.3 | 35.2 | 29.5 | 40.2 | 27.1 | 44.5 | 22.5 | 58.7 | 36.3 | |
| BGE-M3System Type=text-based RAG2026.04 | 39.3 | 32 | 20.8 | 21.7 | 40.3 | 14.7 | 35.4 | 18.5 | 84.3 | 34.8 | |
| Pixel-ReasonerCategory=Tool-Enhanced VLMs2026.01 | 39.28 | 42.09 | 38.14 | 36.52 | 26.73 | 40.2 | — | — | — | — | |
| DocDancerModel=Qwen3-VL-30B-A3B-Thinking2026.06 | 39.2 | — | — | — | — | — | — | — | — | 36.4 | |
| GLM-4.1V + CDSParams=9B, Retriever=CDS2026.06 | 39.15 | — | — | — | — | — | — | — | — | — | |
| M3doc RAGSystem Type=agent-based, VLM=Qwen2.5-VL-7B, LLM=Qwen3-8B2026.04 | 38.4 | 39.2 | 26.7 | 29.8 | 39 | 32 | 50.3 | 21.2 | 40.7 | 36.7 | |
| MRAG^txt+img_ColPaliRetriever=ColPali, Generator=LVLM, Modality (TXT)=true, Modality (IMG)=true2026.02 | 38.1 | 34.1 | 29.4 | 28.1 | 25.7 | 25.3 | 35.8 | 20.3 | 71.1 | — | |
| Doc-ReactModel=GPT-4o2026.06 | 38.1 | — | — | — | — | — | — | — | — | 38.3 | |
| MM-Doc-R1Backbone=Qwen3-4B, Reinforcement Learning Algorithm=None2026.04 | 37.7 | 28.9 | 23.8 | 23.1 | 35.3 | 22.4 | 37.1 | 18.6 | 72.2 | 32.2 | |
| OpenVLThinkerCategory=Vision-Language Reasoning Models (VLRMs)2026.01 | 37.49 | 41.61 | 38.14 | 33.71 | 28.57 | 35.88 | — | — | — | — | |
| VRAG^page_CLIPRetriever=CLIP, Generator=LVLM, Modality (TXT)=false, Modality (IMG)=true2026.02 | 37.3 | 28.9 | 27.7 | 24.7 | 22.5 | 28.3 | 34.8 | 16.7 | 75.4 | — | |
| LC^pageRetriever=-, Generator=LVLM, Modality (TXT)=false, Modality (IMG)=true2026.02 | 37.2 | 41 | 26.9 | 31.5 | 30.3 | 27.6 | 39.7 | 23.6 | 53.1 | — | |
| COAModel=Gemini 2.0 Flash2026.06 | 37.2 | — | — | — | — | — | — | — | — | 31.9 | |
| LC^txtRetriever=-, Generator=LLM, Modality (TXT)=true, Modality (IMG)=false2026.02 | 36.9 | 44.3 | 31.9 | 20.8 | 12.8 | 22 | 32.8 | 21.9 | 69.3 | — | |
| RAG^txt_BM25Retriever=BM25, Generator=LLM, Modality (TXT)=true, Modality (IMG)=false2026.02 | 36.8 | 44.9 | 26.1 | 20.2 | 14.2 | 17.1 | 31.6 | 18.9 | 76.3 | — | |
| MRAG^txt+img_CLIPRetriever=CLIP, Generator=LVLM, Modality (TXT)=true, Modality (IMG)=true2026.02 | 36.7 | 39 | 27.7 | 27 | 13.8 | 21.4 | 34 | 19.2 | 70.2 | — | |
| ColqwenSystem Type=multi-modal RAG, Backbone=ColQwen2.5-7B-multilingual2026.04 | 36.5 | 27.8 | 25 | 16.5 | 22.4 | 23.7 | 33.9 | 13.7 | 82.5 | 31.2 | |
| BM25System Type=text-based RAG2026.04 | 36.4 | 30.9 | 23.4 | 22.3 | 28.5 | 9.2 | 30.7 | 14.1 | 88.3 | 31 | |
| R1-OnevisionCategory=Vision-Language Reasoning Models (VLRMs)2026.01 | 36.09 | 37.58 | 33.05 | 32.58 | 27.65 | 36.88 | — | — | — | — | |
| RAG^txt_BGE-m3Retriever=BGE-m3, Generator=LLM, Modality (TXT)=true, Modality (IMG)=false2026.02 | 36 | 40 | 31.1 | 20.2 | 10.6 | 18.4 | 29.2 | 19.4 | 77.2 | — | |
| VRAG-RLCategory=Tool-Enhanced VLMs2026.01 | 35.97 | 36.58 | 33.55 | 39.33 | 33.9 | 25.81 | — | — | — | — | |
| EVisRAGCategory=Multimodal Retrieval-Augmented Generation Models (MRAGs)2026.01 | 35.86 | 37.92 | 38.14 | 32.02 | 27.19 | 33.89 | — | — | — | — | |
| ThinkLite-VLCategory=Vision-Language Reasoning Models (VLRMs)2026.01 | 35.74 | 37.25 | 33.05 | 30.34 | 27.65 | 37.21 | — | — | — | — | |
| GOTCategory=Prompting Methods2026.01 | 35.04 | 37.58 | 35.59 | 32.02 | 25.81 | 33.89 | — | — | — | — | |
| Mdoc agentSystem Type=agent-based, VLM=Qwen2.5-VL-7B, LLM=Qwen3-8B2026.04 | 35 | 33.1 | 29.3 | 25.8 | 32.6 | 30 | 43.7 | 18.4 | 43.4 | 33.3 | |
| VisionMattersCategory=Vision-Language Reasoning Models (VLRMs)2026.01 | 34.58 | 37.25 | 33.9 | 33.15 | 24.88 | 33.22 | — | — | — | — | |
| MM-Search-R1Category=Multimodal Retrieval-Augmented Generation Models (MRAGs)2026.01 | 33.99 | 35.57 | 33.05 | 34.27 | 23.96 | 33.89 | — | — | — | — | |
| TOTCategory=Prompting Methods2026.01 | 33.88 | 34.23 | 36.44 | 30.9 | 23.96 | 33.89 | — | — | — | — | |
| DirectCategory=Prompting Methods2026.01 | 32.36 | 35.91 | 30.51 | 33.15 | 25.35 | 29.9 | — | — | — | — | |
| Qwen2.5-VL-7B + CDSParams=7B, Retriever=CDS2026.06 | 32.3 | — | — | — | — | — | — | — | — | — | |
| MRAG^txt+img_SigLIPRetriever=SigLIP, Generator=LVLM, Modality (TXT)=true, Modality (IMG)=true2026.02 | 32.2 | 23.9 | 15.1 | 16.9 | 15.1 | 18.1 | 21.9 | 14.7 | 82 | — | |
| M3DocRAGModel=Qwen2-VL-7B2026.06 | 31.4 | — | — | — | — | — | — | — | — | 36.5 | |
| VisRAGModel=GPT-4o2026.06 | 29 | — | — | — | — | — | — | — | — | 27.8 | |
| DocopilotModel=InternVL2-8B2026.06 | 28.8 | — | — | — | — | — | — | — | — | 23 | |
| Qwen2.5-VL-7BParams=7B, Retriever=None (Single-VLM)2026.06 | 28.36 | — | — | — | — | — | — | — | — | — | |
| VisDomCategory=Multimodal Retrieval-Augmented Generation Models (MRAGs)2026.01 | 26.78 | 29.87 | 27.12 | 24.72 | 20.28 | 21.59 | — | — | — | — | |
| VRAG^page_SigLIPRetriever=SigLIP, Generator=LVLM, Modality (TXT)=false, Modality (IMG)=true2026.02 | 26.3 | 16.1 | 13.5 | 13.5 | 9.2 | 9.5 | 15.4 | 9.2 | 77.2 | — | |
| SV-RAGModel=GPT-4o2026.06 | 23 | — | — | — | — | — | — | — | — | 24.2 |