Multi-hop Question Answering on Average (HotpotQA, 2WikiMultiHopQA, MusiQue) (Cost/Efficiency)
0.14Cost per PassStandard RAG
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Standard RAG2026.06 | 0.14 | 37.86 | 0.04 | 1,077 | |
| IRCoT2026.06 | 0.17 | 44.56 | 0.06 | 1,284 | |
| Search-o12026.06 | 0.49 | 50.73 | 0.21 | 4,095 | |
| GDP-RAG2026.06 | 0.51 | 60.63 | 0.27 | 5,717 | |
| PAR-RAG2026.06 | 0.65 | 58.2 | 0.34 | 7,112 | |
| Godbole et al.adaptation=QA first planning stage only2026.06 | 1.12 | 46.28 | 0.45 | 9,606 | |
| KnowTracemode=inference-time KG-tracing, self-bootstrapping fine-tuning=False2026.06 | 1.57 | 51.65 | 0.61 | 13,519 |