Multi-hop Question Answering on MultiHop-RAG
88.36Overall AccuracySPA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SPAModel=Meta-Llama-3-8B, Generator=GPT-4o-mini, Number of Tokens=15M2026.03 | 88.36 | — | — | — | — | |
| SPAModel=Qwen2.5-7B, Generator=GPT-4o-mini, Number of Tokens=15M2026.03 | 86.64 | — | — | — | — | |
| EntiGraphModel=Qwen2.5-7B, Generator=GPT-4o-mini, Number of Tokens=15M2026.03 | 85.42 | — | — | — | — | |
| EntiGraphModel=Meta-Llama-3-8B, Generator=GPT-4o-mini, Number of Tokens=15M2026.03 | 84.31 | — | — | — | — | |
| Active ReadingModel=Qwen2.5-7B, Generator=GPT-4o-mini, Number of Tokens=15M2026.03 | 79.9 | — | — | — | — | |
| Active ReadingModel=Meta-Llama-3-8B, Generator=GPT-4o-mini, Number of Tokens=15M2026.03 | 78.68 | — | — | — | — | |
| Cross-Encoder RerankerGeneration LLM=gpt-4o-mini2026.06 | 77.78 | 96.69 | 66 | 94.02 | 60.21 | |
| Ricci-FiltrationGeneration LLM=gpt-4o-mini2026.06 | 76.56 | 97.18 | 64.02 | 95 | 56.6 | |
| LLM-based RerankerGeneration LLM=gpt-4o-mini2026.06 | 75.59 | 95.22 | 63.67 | 93.02 | 56.6 | |
| No RerankerGeneration LLM=gpt-4o-mini2026.06 | 75.43 | 94.61 | 63.79 | 93.02 | 56.6 | |
| BaseModel=Meta-Llama-3-8B, Generator=GPT-4o-mini, Number of Tokens=15M2026.03 | 73.16 | — | — | — | — | |
| Ricci-FiltrationGeneration LLM=Llama 3.1-8B Instruct2026.06 | 73.12 | 93.87 | 58.53 | 83.39 | 60.21 | |
| HippoRAG2Backbone=Llama-3.1-8B-Instruct2025.02 | 70.27 | 91.54 | 58.41 | 85.71 | 49.91 | |
| Community-GraphRAG (Local)Backbone=Llama-3.1-8B-Instruct2025.02 | 69.01 | 86.89 | 60.63 | 80.07 | 50.6 | |
| RaptorRAGBackbone=Llama-3.1-8B-Instruct2025.02 | 68.78 | 91.91 | 55.26 | 90.03 | 45.28 | |
| Cross-Encoder RerankerGeneration LLM=Llama 3.1-8B Instruct2026.06 | 68.27 | 90.93 | 56.66 | 60.8 | 57.46 | |
| LLM-based RerankerGeneration LLM=Llama 3.1-8B Instruct2026.06 | 67.33 | 90.32 | 54.09 | 56.81 | 60.03 | |
| RAGBackbone=Llama-3.1-8B-Instruct2025.02 | 67.02 | 92.16 | 57.59 | 96.01 | 30.7 | |
| No RerankerGeneration LLM=Llama 3.1-8B Instruct2026.06 | 66.93 | 87.62 | 55.96 | 57.81 | 58.83 | |
| Community-GraphRAG (Global)Backbone=Llama-3.1-8B-Instruct2025.02 | 64.4 | 89.34 | 64.02 | 19.27 | 53.34 | |
| BaseModel=Qwen2.5-7B, Generator=GPT-4o-mini, Number of Tokens=15M2026.03 | 60.91 | — | — | — | — | |
| KG-GraphRAG (Triplets+Text)Backbone=Llama-3.1-8B-Instruct2025.02 | 48.51 | 67.4 | 34.7 | 97.34 | 17.15 | |
| KG-GraphRAG (Triplets only)Backbone=Llama-3.1-8B-Instruct2025.02 | 41.24 | 55.76 | 22.55 | 98.67 | 18.7 |