List your company

Multi-hop Question Answering on Average (MuSiQue, 2WikiMQA, HotpotQA)

43.3EMGraph + LLM
10.5419.04527.5536.055Oct 22, 2024Jan 26, 2025May 2, 2025Aug 7, 2025Nov 11, 2025Feb 15, 2026May 23, 2026

Evaluation results

MethodLinks
Graph + LLMKG Source=D-SynthKG-8b2024.1043.356.9————
Graph + LLMKG Source=Llama-3-70b2024.1041.354.8————
HippoRAGKG Source=GPT-4o2024.1040.354.6————
Dense + LLMKG Source=None2024.1039.453.6————
GraphRAG (local)KG Source=GPT-4o2024.103949.1————
GraphRAG (drift)KG Source=GPT-4o2024.1038.451.3————
R1-SearcherBackbone=7B-Base, Train mix=HQA+2Wiki2026.0537.1——44.251.315.8
LlamaIndexKG Source=D-SynthKG-8b2024.1036.748.5————
Dense RetrieverKG Source=None2024.1036.350.5————
Chain-of-TripletKG Source=D-SynthKG-8b2024.1035.449————
Typed+Shape (ours)Backbone=7B-Inst, Train mix=NQ+HQA2026.0535.4——39.141.526.7
LlamaIndexKG Source=Llama-3-70b2024.1034.745.9————
Typed BaseBackbone=7B-Inst, Train mix=NQ+HQA2026.0534.6——37.840.525.5
GlobalRAGBackbone=7B-Inst, Train mix=2Wiki+HQA2026.0533.6——38.247.814.9
Binary GapBackbone=7B-Inst, Train mix=NQ+HQA2026.0533.4——36.63924.6
Search-R1 (PPO)Backbone=7B-Inst, Train mix=NQ+HQA2026.0531——3741.414.6
LlamaIndexKG Source=Llama-3-8b2024.1030.842.5————
Chain-of-TripletKG Source=Llama-3-70b2024.1030.341.5————
Search-R1 (GRPO)Backbone=7B-Inst, Train mix=NQ+HQA2026.0529.8——38.634.616.2
Search-OnlyBackbone=7B-Inst, Train mix=NQ+HQA2026.0529.8——38.634.616.2
StepSearchBackbone=7B-Inst, Train mix=MuSiQue2026.0528.2——34.836.313.4
Graph + LLMKG Source=Llama-3-8b2024.1028.140.2————
Chain-of-TripletKG Source=Llama-3-8b2024.1023.836.5————
Naive RAGBackbone=7B-Inst2026.0519.7——29.923.55.8
NoneKG Source=None, Retrieval=None2024.1019.333.3————
SFTBackbone=7B-Inst, Train mix=NQ+HQA2026.0518.1——21.725.96.6
IRCoTBackbone=7B-Inst2026.0511.8——13.314.97.2
BubbleRAGModel Scale=30B & 8B2026.03—63.0266.63———
Clue-RAGModel Scale=30B & 8B2026.03—43.2948.65———
CoT + LLMModel Scale=30B & 8B2026.03—14.5123.35———
HippoRAG2Model Scale=30B & 8B2026.03—60.564.4———
LightRAG (global)Model Scale=30B & 8B2026.03—25.5627.17———
LightRAG (hybrid)Model Scale=30B & 8B2026.03—47.7449.92———
LightRAG (local)Model Scale=30B & 8B2026.03—50.7352.94———
Naive RAGModel Scale=30B & 8B2026.03—53.5156.71———
RAPTORModel Scale=30B & 8B2026.03—44.746.97———
ToGModel Scale=30B & 8B2026.03—24.1530.16———
Vanilla LLMModel Scale=30B & 8B2026.03—12.9723.88———