General QA on PopQA
52Exact Match (EM)IRCoT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IRCoTBackbone=Qwen2.5-72B-Instruct2026.03 | 52 | 64.5 | |
| ReSeekBackbone=Qwen2.5-7b-Instruct2025.10 | 50.1 | — | |
| ZeroSearchBackbone=Qwen2.5-7b-Instruct2025.10 | 48.8 | — | |
| ZeroSearchBackbone=Qwen2.5-3b-Instruct2025.10 | 44.8 | — | |
| ReSeekBackbone=Qwen2.5-3b-Instruct2025.10 | 43.4 | — | |
| HippoRAGBackbone=Qwen2.5-72B-Instruct2026.03 | 41.1 | 49.7 | |
| Search-R1Backbone=Qwen2.5-7b-Instruct2025.10 | 39.7 | — | |
| RAGBackbone=Qwen2.5-7b-Instruct2025.10 | 39.2 | — | |
| RAGBackbone=Qwen2.5-3b-Instruct2025.10 | 38.7 | — | |
| Search-R1Backbone=Qwen2.5-3b-Instruct2025.10 | 37.8 | — | |
| TaSR-RAGBackbone=Qwen2.5-72B-Instruct2026.03 | 37.7 | 43.2 | |
| GraphRAG-RouterSystem=Our Method2026.03 | 36.8 | — | |
| GraphRAGBackbone=Qwen2.5-72B-Instruct2026.03 | 36.7 | 45.8 | |
| StructRAGBackbone=Qwen2.5-72B-Instruct2026.03 | 36.5 | 45 | |
| TaSR-RAGBackbone=Qwen2.5-7B-Instruct2026.03 | 36.5 | 43.9 | |
| HyperGraphRAGBackbone=Qwen2.5-72B-Instruct2026.03 | 36 | 45.9 | |
| Router-R1System=RL-based Agent2026.03 | 35.1 | — | |
| RAGBackbone=Qwen2.5-72B-Instruct2026.03 | 33.2 | 41.2 | |
| CoTSystem=Basic LLM2026.03 | 32.7 | — | |
| GraphSearchSystem=Training-free Agent2026.03 | 32.4 | — | |
| StructRAGBackbone=Qwen2.5-7B-Instruct2026.03 | 31.4 | 40.1 | |
| Search-o1System=Training-free Agent2026.03 | 30.2 | — | |
| IRCoTBackbone=Qwen2.5-7B-Instruct2026.03 | 30.1 | 60.2 | |
| GraphRAGSystem=RAG-Based2026.03 | 29.1 | — | |
| CoT ReasoningBackbone=GPT-4o-mini2025.11 | 28.13 | 32.66 | |
| Prompt-R1Backbone=GPT-4o-mini2025.11 | 28.13 | 33.77 | |
| Graph-R1System=RL-based Agent2026.03 | 27.4 | — | |
| GEPAOptimization Framework=APO, Backbone=GPT-4o-mini2025.11 | 27.34 | 31.18 | |
| HyperGraphRAGSystem=RAG-Based2026.03 | 27.2 | — | |
| LinearRAGSystem=RAG-Based2026.03 | 26.9 | — | |
| HippoRAGBackbone=Qwen2.5-7B-Instruct2026.03 | 26.8 | 37 | |
| HippoRAG2System=RAG-Based2026.03 | 26.6 | — | |
| Search-o1Backbone=Qwen2.5-3b-Instruct2025.10 | 26.2 | — | |
| RAPTORSystem=RAG-Based2026.03 | 25.8 | — | |
| BaselineBackbone=GPT-4o-mini2025.11 | 25.78 | 30.75 | |
| Search-R1System=RL-based Agent2026.03 | 25.6 | — | |
| CoTBackbone=Qwen2.5-72B-Instruct2026.03 | 24.4 | 30.3 | |
| Vanilla RAGSystem=RAG-Based2026.03 | 24.2 | — | |
| OPROOptimization Framework=APO, Backbone=GPT-4o-mini2025.11 | 23.44 | 30.14 | |
| R1Backbone=Qwen2.5-7b-Instruct2025.10 | 19.9 | — | |
| Direct InferenceBackbone=Qwen2.5-72B-Instruct2026.03 | 19.7 | 24.4 | |
| TextGradOptimization Framework=APO, Backbone=GPT-4o-mini2025.11 | 19.53 | 28.28 | |
| RAGBackbone=Qwen2.5-7B-Instruct2026.03 | 19.2 | 44.3 | |
| R1Backbone=Qwen2.5-3b-Instruct2025.10 | 17.1 | — | |
| Direct InferenceBackbone=Qwen2.5-7B-Instruct2026.03 | 14 | 16.4 | |
| Direct InferenceBackbone=Qwen2.5-7b-Instruct2025.10 | 14 | — | |
| Search-o1Backbone=Qwen2.5-7b-Instruct2025.10 | 13.1 | — | |
| GraphRAGBackbone=Qwen2.5-7B-Instruct2026.03 | 12.8 | 27.4 | |
| Direct InferSystem=Basic LLM2026.03 | 11.7 | — | |
| Direct InferenceBackbone=Qwen2.5-3b-Instruct2025.10 | 10.8 | — | |
| SFTBackbone=Qwen3-4B2025.11 | 7.81 | 9.53 | |
| GRPOBackbone=Qwen3-4B2025.11 | 7.81 | 12.27 | |
| BaselineBackbone=Qwen3-4B2025.11 | 7.03 | 9.97 | |
| CoT ReasoningBackbone=Qwen3-4B2025.11 | 7.03 | 9.58 | |
| CoTBackbone=Qwen2.5-7B-Instruct2026.03 | 5.4 | 16.4 | |
| CoTBackbone=Qwen2.5-7b-Instruct2025.10 | 5.4 | — | |
| HypergraphRAGBackbone=Qwen2.5-7B-Instruct2026.03 | 0.8 | 1.6 | |
| CoTBackbone=Qwen2.5-3b-Instruct2025.10 | 0.5 | — |