Document Retrieval on MMLongBench-Doc Single-page (SP)
75.77Top-1 AccuracyDocArena
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DocArenaPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 75.77 | 85.22 | 2.46 | |
| ReSearchPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 70.64 | 82.96 | 31.25 | |
| Search-R1Policy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 69.82 | 79.67 | 14.68 | |
| IKEAPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 69.82 | 81.52 | 19.85 | |
| DeepResearcherPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 69.61 | 81.93 | 38.84 | |
| AutoRefinePolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 65.71 | 76.8 | 14.81 | |
| StepSearchPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 63.86 | 77 | 3.82 | |
| R-SearchPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 63.45 | 75.15 | 21.28 | |
| ZeroSearchPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 62.63 | 76.18 | 31.03 | |
| ASearcherPolicy Model=Qwen2.5-7B-Instruct, Max Search Turns=2, Retriever=ColPali2026.05 | 56.06 | 63.24 | 41.99 |