List your company

Multi-Hop Question Answering on HotpotQA (SubEM, F1)

58.9F1GraphRAG-Router
21.428831.156940.88550.6131Jan 6, 2026Jan 20, 2026Feb 3, 2026Feb 17, 2026Mar 3, 2026Mar 17, 2026Mar 31, 2026

Evaluation results

MethodLinks
GraphRAG-RouterSystem=Our Method2026.0358.9—
HippoRAG2System=RAG-Based2026.0350.1—
LinearRAGSystem=RAG-Based2026.0348.7—
Graph-R1System=RL-based Agent2026.0346.8—
Router-R1System=RL-based Agent2026.0346.7—
Vanilla RAGSystem=RAG-Based2026.0344.5—
Search-o1System=Training-free Agent2026.0343.9—
CoTSystem=Basic LLM2026.0343.6—
Stable-RAGBackbone=Qwen3-8B, Retriever=Contriever, exhaustive decoding=true2026.0142.9436.89
GraphSearchSystem=Training-free Agent2026.0342.5—
Stable-RAGBackbone=Qwen3-8B, Retriever=Contriever, exhaustive decoding=false2026.0141.7835.73
RAPTORSystem=RAG-Based2026.0340.9—
ATMBackbone=Qwen3-8B, Retriever=Contriever2026.0140.6935.12
Stable-RAGBackbone=LLaMA3-8B-Instruct, Retriever=Contriever, exhaustive decoding=true2026.0140.1639.12
Stable-RAGBackbone=LLaMA3-8B-Instruct, Retriever=Contriever, exhaustive decoding=false2026.0139.8738.91
RetRobustBackbone=Qwen3-8B, Retriever=Contriever2026.0139.3932.77
Pos2DistillBackbone=Qwen3-8B, Retriever=Contriever2026.0139.1133.72
RAATBackbone=Qwen3-8B, Retriever=Contriever2026.0139.0633.54
GraphRAGSystem=RAG-Based2026.0338.8—
Vanilla RAGBackbone=Qwen3-8B, Retriever=Contriever2026.0138.6633.14
Ms-PoEBackbone=Qwen3-8B, Retriever=Contriever2026.0138.2132.98
HyperGraphRAGSystem=RAG-Based2026.0337.6—
Search-R1System=RL-based Agent2026.0337.1—
ATMBackbone=LLaMA3-8B-Instruct, Retriever=Contriever2026.0136.9734.36
RAATBackbone=LLaMA3-8B-Instruct, Retriever=Contriever2026.0136.1233.58
Direct InferSystem=Basic LLM2026.0336.1—
Pos2DistillBackbone=LLaMA3-8B-Instruct, Retriever=Contriever2026.0135.7932.73
Stable-RAGBackbone=Qwen3-8B, Retriever=DPR, exhaustive decoding=true2026.0135.7831.77
RetRobustBackbone=LLaMA3-8B-Instruct, Retriever=Contriever2026.0135.3431.46
Vanilla SFTBackbone=Qwen3-8B, Retriever=Contriever2026.0134.1528.36
Vanilla RAGBackbone=LLaMA3-8B-Instruct, Retriever=Contriever2026.0134.0830.73
RAATBackbone=Qwen3-8B, Retriever=DPR2026.0133.7527.21
Ms-PoEBackbone=LLaMA3-8B-Instruct, Retriever=Contriever2026.0133.6530.17
ATMBackbone=Qwen3-8B, Retriever=DPR2026.0133.4329.07
Stable-RAGBackbone=Qwen3-8B, Retriever=DPR, exhaustive decoding=false2026.0133.2630.15
RetRobustBackbone=Qwen3-8B, Retriever=DPR2026.0133.0626.83
Stable-RAGBackbone=LLaMA3-8B-Instruct, Retriever=DPR, exhaustive decoding=true2026.0132.1230.41
Pos2DistillBackbone=Qwen3-8B, Retriever=DPR2026.0131.8826.53
Stable-RAGBackbone=LLaMA3-8B-Instruct, Retriever=DPR, exhaustive decoding=false2026.0131.6829.48
Vanilla SFTBackbone=LLaMA3-8B-Instruct, Retriever=Contriever2026.0131.5827.25
Vanilla RAGBackbone=Qwen3-8B, Retriever=DPR2026.0131.3326.17
Ms-PoEBackbone=Qwen3-8B, Retriever=DPR2026.0131.0225.93
RetRobustBackbone=LLaMA3-8B-Instruct, Retriever=DPR2026.0130.3626.96
Vanilla SFTBackbone=LLaMA3-8B-Instruct, Retriever=DPR2026.0129.8524.63
Vanilla SFTBackbone=Qwen3-8B, Retriever=DPR2026.0129.7725.35
ATMBackbone=LLaMA3-8B-Instruct, Retriever=DPR2026.0129.3128.55
Pos2DistillBackbone=LLaMA3-8B-Instruct, Retriever=DPR2026.0128.9126.45
RAATBackbone=LLaMA3-8B-Instruct, Retriever=DPR2026.0128.7926.35
Ms-PoEBackbone=LLaMA3-8B-Instruct, Retriever=DPR2026.0128.5726.12
Vanilla RAGBackbone=LLaMA3-8B-Instruct, Retriever=DPR2026.0128.2225.66
Naive GenerationBackbone=Qwen3-8B, Retriever=Contriever2026.0124.8619.54
Naive GenerationBackbone=Qwen3-8B, Retriever=DPR2026.0124.8619.54
Direct GenerationBackbone=LLaMA3-8B-Instruct, Retriever=Contriever2026.0122.8721.39
Direct GenerationBackbone=LLaMA3-8B-Instruct, Retriever=DPR2026.0122.8721.39