Long-context document understanding on MMLongBench-Doc
55.8AccuracySynthetic Reasoning
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Synthetic ReasoningModel family=Qwen3 VL, Scale=32B, Strategy=Synthetic Reasoning2026.03 | 55.8 | — | |
| Plain DistillationModel family=Qwen3 VL, Strategy=Plain Distillation2026.03 | 54.9 | — | |
| Qwen3 VL 235B A22B InstructModel family=Qwen3 VL, Scale=235B A22B, Strategy=Instruct2026.03 | 54.8 | — | |
| LongPOModel family=Qwen3 VL, Strategy=LongPO2026.03 | 53.6 | — | |
| No-thinkModel family=Qwen3 VL, Strategy=No-think2026.03 | 52.2 | — | |
| Qwen3 VL 32B InstructModel family=Qwen3 VL, Scale=32B, Strategy=Instruct2026.03 | 51.8 | — | |
| Synthetic ReasoningModel family=Mistral, Scale=24B, Strategy=Synthetic Reasoning2026.03 | 47.5 | — | |
| Plain DistillationModel family=Mistral, Strategy=Plain Distillation2026.03 | 46.6 | — | |
| GPT-4.1Backbone=-, Param=-, Paradigm=E2E2026.04 | 45.6 | — | |
| Qwen Thinking TracesModel family=Mistral, Strategy=Thinking Traces2026.03 | 45 | — | |
| No-thinkModel family=Mistral, Strategy=No-think2026.03 | 43 | — | |
| GPT-4ocategory=Proprietary2024.10 | 42.9 | 44.9 | |
| GPT-4oVenue=-, Param.=-, Evi.=P, RAG=×2026.04 | 42.8 | — | |
| GPT-4oBackbone=-, Param=-, Paradigm=E2E2026.04 | 42.8 | — | |
| Doc-V* (GRPO)Backbone=Qwen2.5-VL, Param=7B, Paradigm=Agent2026.04 | 42.1 | — | |
| MoLoRAGBackbone=Qwen2.5-VL, Param=7B, Paradigm=RAG2026.04 | 41 | — | |
| DocSeekerVenue=-, Param.=7B, Evi.=P, RAG=×2026.04 | 40.1 | — | |
| Mistral 3.1 Small 24BModel family=Mistral, Scale=24B2026.03 | 39.9 | — | |
| Doc-V* (SFT)Backbone=Qwen2.5-VL, Param=7B, Paradigm=Agent2026.04 | 39.8 | — | |
| DocSeeker-SFTVenue=-, Param.=7B, Evi.=P, RAG=×2026.04 | 38.6 | — | |
| Qwen2.5-VL (RAG Top-5)Backbone=Qwen2.5-VL, Param=7B, Paradigm=RAG2026.04 | 36.1 | — | |
| Claude-3.7-SonnetBackbone=-, Param=-, Paradigm=E2E2026.04 | 33.9 | — | |
| URaGBackbone=Qwen2.5-VL, Param=7B, Paradigm=RAG2026.04 | 33.8 | — | |
| Qwen2-VL-72Bactivated_params=72B, category=Open-source2024.10 | 33.3 | 35.7 | |
| CogDocBackbone=Qwen2.5-VL, Param=7B, Paradigm=Agent2026.04 | 33 | — | |
| GPT-4VVenue=-, Param.=-, Evi.=P, RAG=×2026.04 | 32.4 | — | |
| GPT-4o minicategory=Proprietary2024.10 | 29 | 28.6 | |
| DocopilotVenue=CVPR 2025, Param.=8B, Evi.=T+I, RAG=×2026.04 | 28.8 | — | |
| Baseline-SFT (short-answer)Venue=-, Param.=7B, Evi.=P, RAG=×2026.04 | 28.8 | — | |
| DocopilotBackbone=InternVL2, Param=8B, Paradigm=E2E2026.04 | 28.8 | — | |
| GPT-4o miniBackbone=-, Param=-, Paradigm=E2E2026.04 | 28.6 | — | |
| ARIAactivated_params=3.9B (25.3B), category=Open-source2024.10 | 28.3 | 24.6 | |
| Gemini-1.5-Procategory=Proprietary2024.10 | 28.2 | 20.6 | |
| Gemini-1.5-ProVenue=-, Param.=-, Evi.=P, RAG=×2026.04 | 28.2 | — | |
| Gemini-1.5-ProBackbone=-, Param=-, Paradigm=E2E2026.04 | 28.2 | — | |
| Qwen2.5-VL (Baseline)Backbone=Qwen2.5-VL, Param=7B, Paradigm=E2E2026.04 | 28 | — | |
| Gemini-1.5-Flashcategory=Proprietary2024.10 | 27 | 21.3 | |
| VRAG-RLBackbone=Qwen2.5-VL, Param=7B, Paradigm=Agent2026.04 | 26.6 | — | |
| BaselineVenue=-, Param.=7B, Evi.=P, RAG=×2026.04 | 25.4 | — | |
| InternVL3Venue=Arxiv 2025, Param.=8B, Evi.=P, RAG=×2026.04 | 24.1 | — | |
| InternVL3Backbone=InternViT / Qwen2.5, Param=8B, Paradigm=E2E2026.04 | 24.1 | — | |
| SV-RAGVenue=ICLR 2025, Param.=4B, Evi.=P, RAG=✓2026.04 | 23 | — | |
| SV-RAGBackbone=InternVL2, Param=4B, Paradigm=RAG2026.04 | 23 | — | |
| Qwen2-VL-7Bactivated_params=7B, category=Open-source2024.10 | 21.3 | 22.7 | |
| M3DocRAGVenue=Arxiv 2024, Param.=9B, Evi.=P, RAG=✓2026.04 | 21 | — | |
| M3DocRAGBackbone=Qwen2-VL, Param=7B, Paradigm=RAG2026.04 | 21 | — | |
| Vis-RAGVenue=ICLR 2025, Param.=4B, Evi.=P, RAG=✓2026.04 | 18.8 | — | |
| VisRAGBackbone=MiniCPM-V 2.6, Param=8B, Paradigm=RAG2026.04 | 18.8 | — | |
| VDocRAGVenue=CVPR 2025, Param.=8B, Evi.=P, RAG=✓2026.04 | 18.4 | — | |
| VDocRAGBackbone=Phi3-Vision, Param=4B, Paradigm=RAG2026.04 | 18.4 | — | |
| InternVL2-40Bactivated_params=40B, category=Open-source2024.10 | 18.2 | 17.9 | |
| InternVL-Chat-V1.5activated_params=26B, category=Open-source2024.10 | 14.6 | 13 | |
| Llama3.2-11Bactivated_params=11B, category=Open-source2024.10 | 13.8 | 11.3 | |
| mPLUG-DocOwl2Venue=ACL 2025, Param.=8B, Evi.=P, RAG=×2026.04 | 13.4 | — | |
| mPLUG-DocOwl2Backbone=ViT / LLaMa, Param=8B, Paradigm=E2E2026.04 | 13.4 | — | |
| MiniCPM-V-2.6activated_params=8B, category=Open-source2024.10 | 11.5 | 11.6 | |
| Idefics2activated_params=8B, category=Open-source2024.10 | 7 | 6.8 | |
| Pixtral-12Bactivated_params=12B, category=Open-source2024.10 | 6.4 | 6 |