Question Answering on MetaQA 3-hop (Acc/F1) benchmark leaderboard · SOTA2 Research

Question Answering on MetaQA 3-hop (Acc/F1)

93.4AccuracyRELOOP

Evaluation Results

Method
RELOOP2025.10	93.4	88.3
RELOOP2025.10	90.1	86.6
AdaptiveRAG2025.10	84.5	85.7
Graph-constrained Reasoning (GcR)2025.10	83.2	80.6
Think on Graph (ToG)2025.10	81.1	78.5
Falcon-H1-7B-Instruct2025.10	44.6	—
Falcon3-10B-instruct2025.10	39.8	—
Qwen3-4B-Instruct-25072025.10	38.5	—
Llama-3.1-8B-Instruct2025.10	31.9	—
Graph-S^32025.10	14.7	36.29
Graph-S^32025.10	12.73	29.49
Graph-S^32025.10	10.73	29.55
RAG/1hop2025.10	4.13	11.99
KG-Agent2025.10	4.07	10.89
RAG/1hop2025.10	4.03	15.46
KG-Agent2025.10	3.85	8.29
RAG/1hop2025.10	3.3	18.61
RAG/2hop2025.10	3.07	14.04
KG-Agent2025.10	2.64	7.2
RAG/2hop2025.10	2.6	9.42
RAG/2hop2025.10	1.8	11.3
ToG2025.10	1.43	6.1
No graph2025.10	1.23	8.86
LightRAG2025.10	1	6.38
LightRAG2025.10	0.53	4.28
G-retriever2025.10	0.4	3.31
No graph2025.10	0.33	4.6
No graph2025.10	0.2	1.19
RAG/3hop2025.10	0.13	3.45
RAG/3hop2025.10	0.13	4.23
G-retriever2025.10	0.1	1.77
ToG2025.10	0.1	2.8
LightRAG2025.10	0.07	0.4
RAG/3hop2025.10	0.07	5.76
G-retriever2025.10	0.03	1.87
No retriever2025.10	0	1.25
ToG2025.10	0	0.2
No retriever2025.10	0	2.94
No retriever2025.10	0	1.05