Multi-hop Retrieval on HotPotQA (Including Efficiency Metrics)
0.065Latency (s)Search-R1-7B
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Search-R1-7BModel Scale=7B2025.07 | 0.065 | 48.2 | 2,119 | 2.99 | 1.28 | |
| O2 Searcher-3BModel Scale=3B2025.07 | 0.105 | 50.1 | 2,937 | 1.79 | 1.77 | |
| MDRRetrieval Strategy=Reasoning-Augmented Dense Retrieval2026.05 | 0.13 | — | — | — | — | |
| R1 Searcher-7BModel Scale=7B2025.07 | 0.1335 | 69.1 | 4,458 | 6.17 | 2.22 | |
| CoRAG-8BModel Scale=8B2025.07 | 0.1486 | 64.3 | 7,770 | 1.24 | 4 | |
| FrugalRAG-7BModel Scale=7B2025.07 | 0.2415 | 70.4 | 9,138 | 1.27 | 2.89 | |
| SimpleDeepSearcher-7BModel Scale=7B2025.07 | 0.6571 | 64.8 | 9,657 | 1.31 | 2.75 | |
| R3-RAGRetrieval Strategy=LLM-Driven Multi-Step Retrieval2026.05 | 0.93 | — | — | — | — | |
| ThinkGRRetrieval Strategy=Our Method2026.05 | 1.13 | — | — | — | — | |
| Auto-RAGRetrieval Strategy=LLM-Driven Multi-Step Retrieval2026.05 | 4.37 | — | — | — | — | |
| IRCoTRetrieval Strategy=LLM-Driven Multi-Step Retrieval2026.05 | 9.26 | — | — | — | — | |
| SelfaskRetrieval Strategy=LLM-Driven Multi-Step Retrieval2026.05 | 37.81 | — | — | — | — | |
| ITER-RETGENRetrieval Strategy=LLM-Driven Multi-Step Retrieval2026.05 | 50.87 | — | — | — | — | |
| RT-RAGRetrieval Strategy=LLM-Driven Multi-Step Retrieval2026.05 | 52.53 | — | — | — | — | |
| GritHopperRetrieval Strategy=Reasoning-Augmented Dense Retrieval2026.05 | 65.51 | — | — | — | — |