Multi-hop Question Answering on NQ
25.5F1 (token-overlap)Oracle
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Oracletype=upper bound2026.04 | 25.5 | — | |
| GraphRAG2026.04 | 24.2 | — | |
| LR+DirectGR2026.04 | 20.1 | 71.1 | |
| ReasonRAGmode=lex2026.04 | 10.1 | — | |
| HybridRouter2026.04 | 7.4 | — | |
| VanillaRAG2026.04 | 7 | — |