Multi-hop Question Answering on 2WikiQA (test)
74.3F1MultiCube-RAG
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MultiCube-RAGGenerator=Llama3.3-70B-Instruct2026.02 | 74.3 | 66.8 | — | |
| HippoRAGGenerator=Llama3.3-70B-Instruct2026.02 | 71.8 | 65 | — | |
| MultiCube-RAGGenerator=GPT-4o-mini2026.02 | 71.5 | 63.2 | — | |
| HippoRAG 2Generator=Llama3.3-70B-Instruct2026.02 | 71 | 65 | — | |
| HippoRAG 2Generator=GPT-4o-mini2026.02 | 69.7 | 60.5 | — | |
| HippoRAGGenerator=GPT-4o-mini2026.02 | 67.3 | 59.4 | — | |
| HERA (Qwen-3-14B)category=Ours2026.04 | 64.77 | 59.5 | 60.02 | |
| DeepResearcher-7Bcategory=Advanced RAG2026.04 | 64.05 | 54.35 | 49.4 | |
| RankCoTGenerator=GPT-4o-mini2026.02 | 62.7 | 52.5 | — | |
| IRCoTGenerator=GPT-4o-mini2026.02 | 62.6 | 54.8 | — | |
| NV-Embed-v2 (7B)Generator=Llama3.3-70B-Instruct2026.02 | 61.5 | 57.5 | — | |
| GraphRAGGenerator=GPT-4o-mini2026.02 | 61 | 45.7 | — | |
| NV-Embed-v2 (7B)Generator=GPT-4o-mini2026.02 | 60.8 | 54.4 | — | |
| GraphRAGGenerator=Llama3.3-70B-Instruct2026.02 | 58.6 | 51.4 | — | |
| HippoRAG2 + DeepRefine-8BConstructor=AR1, Refinement Model=DeepRefine-8B2026.05 | 55.45 | — | — | |
| HERA (Llama-3.1-8)category=Ours2026.04 | 55.42 | 46.7 | 52.62 | |
| HippoRAG2 + DeepRefine-4BConstructor=AR1, Refinement Model=DeepRefine-4B2026.05 | 55.06 | — | — | |
| HippoRAG2 + DeepRefine-8BConstructor=Naive, Refinement Model=DeepRefine-8B2026.05 | 53.02 | — | — | |
| AceSearcher-8Bcategory=Advanced RAG2026.04 | 52.75 | 45.59 | 50.83 | |
| HippoRAG + DeepRefine-8BConstructor=AR1, Refinement Model=DeepRefine-8B2026.05 | 52.75 | — | — | |
| HippoRAG2Constructor=AR12026.05 | 52.61 | — | — | |
| RAPTORGenerator=Llama3.3-70B-Instruct2026.02 | 52.1 | 47.3 | — | |
| HippoRAG + DeepRefine-4BConstructor=AR1, Refinement Model=DeepRefine-4B2026.05 | 51.89 | — | — | |
| BM25Generator=GPT-4o-mini2026.02 | 51.8 | 46.9 | — | |
| ExSearchcategory=Advanced RAG, Backbone=Qwen2.5-7B2026.04 | 51.56 | 44.03 | 44.83 | |
| HippoRAG2Constructor=Naive2026.05 | 51.47 | — | — | |
| HippoRAG2 + DeepRefine-4BConstructor=Naive, Refinement Model=DeepRefine-4B2026.05 | 51.26 | — | — | |
| HippoRAGConstructor=AR12026.05 | 50.34 | — | — | |
| HippoRAG + DeepRefine-4BConstructor=Naive, Refinement Model=DeepRefine-4B2026.05 | 50.23 | — | — | |
| ToG + DeepRefine-8BConstructor=AR1, Refinement Model=DeepRefine-8B2026.05 | 50.13 | — | — | |
| ToG + DeepRefine-4BConstructor=AR1, Refinement Model=DeepRefine-4B2026.05 | 50.08 | — | — | |
| HippoRAG + DeepRefine-8BConstructor=Naive, Refinement Model=DeepRefine-8B2026.05 | 49.84 | — | — | |
| HippoRAGConstructor=Naive2026.05 | 49.77 | — | — | |
| Search-r1-7Bcategory=Advanced RAG2026.04 | 49.2 | 44.69 | 48.89 | |
| Iter-RetGencategory=Advanced RAG, Backbone=Qwen2.5-7B2026.04 | 49.12 | 43.54 | 47.71 | |
| ToGConstructor=AR12026.05 | 48.56 | — | — | |
| CORAG-8B (Greedy)category=Advanced RAG2026.04 | 46.62 | 35.43 | 39.09 | |
| R1-Searcher 7Bcategory=Advanced RAG2026.04 | 45.19 | 35.38 | 40.11 | |
| ToG + DeepRefine-8BConstructor=Naive, Refinement Model=DeepRefine-8B2026.05 | 44.02 | — | — | |
| ToGConstructor=Naive2026.05 | 43.74 | — | — | |
| ToG + DeepRefine-4BConstructor=Naive, Refinement Model=DeepRefine-4B2026.05 | 42.88 | — | — | |
| SELF-RAGcategory=Advanced RAG, Backbone=llama2-7B2026.04 | 42.02 | 37.3 | 40.82 | |
| IRCoTcategory=Advanced RAG, Backbone=GPT-4o-mini2026.04 | 42.01 | 35.5 | 38.73 | |
| BM25Generator=Llama3.3-70B-Instruct2026.02 | 41.9 | 38.1 | — | |
| InstructRAG-8Bcategory=Advanced RAG2026.04 | 40.72 | 35.45 | 36.98 | |
| RAPTORGenerator=GPT-4o-mini2026.02 | 39.7 | 39.7 | — | |
| MMAO-RAG-8Bcategory=Advanced RAG2026.04 | 38.58 | 31.89 | 35.56 | |
| Qwen3-14Bcategory=Single-turn Retrieval2026.04 | 37.79 | 31.1 | 31.57 | |
| Subgraph + DeepRefine-8BConstructor=AR1, Refinement Model=DeepRefine-8B2026.05 | 37.76 | — | — | |
| Subgraph + DeepRefine-4BConstructor=AR1, Refinement Model=DeepRefine-4B2026.05 | 37.21 | — | — | |
| Subgraph + DeepRefine-8BConstructor=Naive, Refinement Model=DeepRefine-8B2026.05 | 35.62 | — | — | |
| SubgraphConstructor=AR12026.05 | 35.6 | — | — | |
| GPT-4o-minicategory=Single-turn Retrieval2026.04 | 35.11 | 28.41 | 28.42 | |
| Subgraph + DeepRefine-4BConstructor=Naive, Refinement Model=DeepRefine-4B2026.05 | 34.64 | — | — | |
| SubgraphConstructor=Naive2026.05 | 33.53 | — | — | |
| HippoRAG2 + DeepRefine-8BConstructor=Graphify (LLM-Wiki), Refinement Model=DeepRefine-8B2026.05 | 32.91 | — | — | |
| Qwen3-8Bcategory=Single-turn Retrieval2026.04 | 32.12 | 26.1 | 26.37 | |
| Qwen2.5-14B-Instructcategory=Single-turn Retrieval2026.04 | 31.85 | 26.25 | 26.64 | |
| Plan-RAGcategory=Advanced RAG, Backbone=8b2026.04 | 31.68 | 27.38 | 30.53 | |
| HippoRAG + DeepRefine-8BConstructor=Graphify (LLM-Wiki), Refinement Model=DeepRefine-8B2026.05 | 31.61 | — | — | |
| HippoRAG + DeepRefine-4BConstructor=Graphify (LLM-Wiki), Refinement Model=DeepRefine-4B2026.05 | 29.84 | — | — | |
| HippoRAGConstructor=Graphify (LLM-Wiki)2026.05 | 29.03 | — | — | |
| HippoRAG2 + DeepRefine-4BConstructor=Graphify (LLM-Wiki), Refinement Model=DeepRefine-4B2026.05 | 28.35 | — | — | |
| HippoRAG2Constructor=Graphify (LLM-Wiki)2026.05 | 27.98 | — | — | |
| Subgraph + DeepRefine-8BConstructor=Graphify (LLM-Wiki), Refinement Model=DeepRefine-8B2026.05 | 26.34 | — | — | |
| Subgraph + DeepRefine-4BConstructor=Graphify (LLM-Wiki), Refinement Model=DeepRefine-4B2026.05 | 25.85 | — | — | |
| SubgraphConstructor=Graphify (LLM-Wiki)2026.05 | 25.37 | — | — | |
| Llama3-3.1-8B-instructcategory=Single-turn Retrieval2026.04 | 25.11 | 20.35 | 20.4 | |
| ToG + DeepRefine-8BConstructor=Graphify (LLM-Wiki), Refinement Model=DeepRefine-8B2026.05 | 21.55 | — | — | |
| ToGConstructor=Graphify (LLM-Wiki)2026.05 | 21.46 | — | — | |
| ToG + DeepRefine-4BConstructor=Graphify (LLM-Wiki), Refinement Model=DeepRefine-4B2026.05 | 21.28 | — | — |