Multi-hop Question Answering on 2Wiki (F1 token-overlap, Gap closed)
57.9F1 (token-overlap)Oracle
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Oracletype=upper bound2026.04 | 57.9 | — | |
| GraphRAG2026.04 | 55.9 | — | |
| LR+DirectGR2026.04 | 48.4 | 72.3 | |
| HybridRouter2026.04 | 39.9 | — | |
| ReasonRAGmode=lex2026.04 | 25.5 | — | |
| VanillaRAG2026.04 | 23.3 | — |