Document Retrieval on MMLongBench-Doc Multi-page (MP)
61.38RecallDocArena
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| DocArenaPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 61.38 | 33.76 | 39.68 | 2.97 | |
| ReSearchPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 56.53 | 33.45 | 38.22 | 45.3 | |
| DeepResearcherPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 55.82 | 33.37 | 38.25 | 44.39 | |
| Search-R1Policy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 55.24 | 32.18 | 37.09 | 22.77 | |
| IKEAPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 54.43 | 34.27 | 38.41 | 23.62 | |
| AutoRefinePolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 52.23 | 32.66 | 36.75 | 15.04 | |
| StepSearchPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 52.09 | 30.31 | 35.24 | 5.3 | |
| R-SearchPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 49.58 | 30.79 | 34.84 | 30.33 | |
| ZeroSearchPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 48.49 | 31.33 | 34.74 | 26.32 | |
| ASearcherPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=3, Retriever=ColPali2026.05 | 45.38 | 25.52 | 29.87 | 46.78 |