Multi-hop Question Answering on HotpotQA l1
0.05Cost per PassStandard RAG
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Standard RAG2026.06 | 0.05 | 56.41 | 0.03 | 696 | |
| IRCoT2026.06 | 0.07 | 54.11 | 0.04 | 810 | |
| GDP-RAG2026.06 | 0.25 | 64.75 | 0.16 | 3,271 | |
| Search-o12026.06 | 0.3 | 51.03 | 0.16 | 2,881 | |
| PAR-RAG2026.06 | 0.39 | 62.79 | 0.24 | 4,856 | |
| Godbole et al.adaptation=QA first planning stage only2026.06 | 0.54 | 59.42 | 0.32 | 6,511 | |
| KnowTracemode=inference-time KG-tracing, self-bootstrapping fine-tuning=False2026.06 | 0.68 | 55.66 | 0.38 | 8,296 |