Long-context Document Understanding on LongDocURL
64.5AccuracyGPT-4o
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-4oBackbone=-, Param=-, Paradigm=E2E2026.04 | 64.5 | |
| Doc-V* (GRPO)Backbone=Qwen2.5-VL, Param=7B, Paradigm=Agent2026.04 | 56.3 | |
| Doc-V* (SFT)Backbone=Qwen2.5-VL, Param=7B, Paradigm=Agent2026.04 | 53 | |
| URaGBackbone=Qwen2.5-VL, Param=7B, Paradigm=RAG2026.04 | 52.2 | |
| MoLoRAGBackbone=Qwen2.5-VL, Param=7B, Paradigm=RAG2026.04 | 51.9 | |
| Gemini-1.5-ProBackbone=-, Param=-, Paradigm=E2E2026.04 | 50.9 | |
| VRAG-RLBackbone=Qwen2.5-VL, Param=7B, Paradigm=Agent2026.04 | 44.9 | |
| VisRAGBackbone=MiniCPM-V 2.6, Param=8B, Paradigm=RAG2026.04 | 41.9 | |
| VDocRAGBackbone=Phi3-Vision, Param=4B, Paradigm=RAG2026.04 | 39.8 | |
| InternVL3Backbone=InternViT / Qwen2.5, Param=8B, Paradigm=E2E2026.04 | 38.7 | |
| Qwen2.5-VL (RAG Top-5)Backbone=Qwen2.5-VL, Param=7B, Paradigm=RAG2026.04 | 37.8 | |
| M3DocRAGBackbone=Qwen2-VL, Param=7B, Paradigm=RAG2026.04 | 35.1 | |
| Qwen2.5-VL (Baseline)Backbone=Qwen2.5-VL, Param=7B, Paradigm=E2E2026.04 | 32.9 | |
| mPLUG-DocOwl2Backbone=ViT / LLaMa, Param=8B, Paradigm=E2E2026.04 | 5.3 |