Single-hop QA on NQ (test)
42Token-level F1RAG-Critic
Evaluation results
| Method | Links | |
|---|---|---|
| RAG-CriticType=Agentic RAG2026.04 | 42 | |
| RePAIRTraining Phase=full2026.04 | 40.3 | |
| MetaRAGType=Agentic RAG2026.04 | 40.2 | |
| RePAIR (online)Training Phase=online2026.04 | 38.8 | |
| RePAIR (offline)Training Phase=offline2026.04 | 38.4 | |
| Standard RAGType=Standard RAG2026.04 | 38.3 | |
| Self-RAGType=Agentic RAG2026.04 | 32.3 | |
| Self-RefineType=Agentic RAG2026.04 | 22.3 | |
| FLAREType=Agentic RAG2026.04 | 19.8 |