Question Answering on MuSiQue (LLM-Acc, Cont-Acc)
74.1LLM AccuracyA-RAG (Full)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| A-RAG (Full)Backbone LLM=GPT-5-mini2026.02 | 74.1 | 65.3 | |
| A-RAG (Naive)Backbone LLM=GPT-5-mini2026.02 | 66.2 | 59.7 | |
| LinearRAGBackbone LLM=GPT-5-mini2026.02 | 62.4 | 51.8 | |
| HippoRAG2Backbone LLM=GPT-5-mini2026.02 | 61.7 | 52.5 | |
| FaithfulRAGBackbone LLM=GPT-5-mini2026.02 | 52.9 | 52.8 | |
| Naive RAGBackbone LLM=GPT-5-mini2026.02 | 52.8 | 48.7 | |
| GraphRAGBackbone LLM=GPT-5-mini2026.02 | 48.3 | 39.1 | |
| A-RAG (Full)Backbone LLM=GPT-4o-mini2026.02 | 46.1 | 39.6 | |
| GAM-RAGMemorization turns=22026.03 | 45.8 | 41.3 | |
| GAM-RAGMemorization turns=12026.03 | 44.5 | 39.2 | |
| A-RAG (Naive)Backbone LLM=GPT-4o-mini2026.02 | 43.8 | 38.5 | |
| GAM-RAGMemorization turns=02026.03 | 43.3 | 38 | |
| HippoRAG2Backbone LLM=GPT-4o-mini2026.02 | 40.6 | 38.4 | |
| MA-RAGBackbone LLM=GPT-5-mini2026.02 | 40 | 31.6 | |
| Naive RAGBackbone LLM=GPT-4o-mini2026.02 | 38.6 | 36.1 | |
| RAGentABackbone LLM=GPT-5-mini2026.02 | 38.3 | 37.4 | |
| LinearRAGMemorization turns=02026.03 | 37.2 | 32.5 | |
| Direct AnswerBackbone LLM=GPT-5-mini2026.02 | 35.8 | 26.5 | |
| REMINDRAGMemorization turns=02026.03 | 35.3 | 30.8 | |
| LinearRAGBackbone LLM=GPT-4o-mini2026.02 | 34.8 | 26.3 | |
| GFM-RAGMemorization turns=02026.03 | 34.2 | 28.7 | |
| MA-RAGBackbone LLM=GPT-4o-mini2026.02 | 34.1 | 27.4 | |
| RAGentABackbone LLM=GPT-4o-mini2026.02 | 32.2 | 29.9 | |
| PoGMemorization turns=02026.03 | 31.6 | 29.2 | |
| all-mpnet-base-v2Memorization turns=02026.03 | 29.6 | 26.8 | |
| nvidia/NV-Embed-v2Memorization turns=02026.03 | 28.9 | 26.1 | |
| FaithfulRAGBackbone LLM=GPT-4o-mini2026.02 | 28.8 | 22.6 | |
| GritLM / GritLM-7BMemorization turns=02026.03 | 28.5 | 25.8 | |
| GraphRAGBackbone LLM=GPT-4o-mini2026.02 | 26.4 | 20.8 | |
| HippoRAG2Memorization turns=02026.03 | 24.8 | 27.3 | |
| DyG-RAGMemorization turns=02026.03 | 20.4 | 17.8 | |
| Direct AnswerBackbone LLM=GPT-4o-mini2026.02 | 18.3 | 13.9 | |
| Chain-of-Thought (CoT)Memorization turns=02026.03 | 15.8 | 13.6 | |
| Direct AnsweringMemorization turns=02026.03 | 14.9 | 13.2 |