Multi-hop Question Answering on HotpotQA (F1 token-overlap, Gap closed)
70.3F1 (Token Overlap)Oracle
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Oracletype=upper bound2026.04 | 70.3 | — | |
| GraphRAG2026.04 | 67.9 | — | |
| LR+DirectGR2026.04 | 55.8 | 79.3 | |
| HybridRouter2026.04 | 52.4 | — | |
| ReasonRAGmode=lex2026.04 | 42.1 | — | |
| VanillaRAG2026.04 | 39.2 | — |