Multi-hop Question Answering on HotpotQA N=1,000
73Judge EMSEAL-RAG
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SEAL-RAGModel=gpt-4o, k=12025.12 | 73 | 91 | 62 | 72 | 14 | |
| SEAL-RAGModel=gpt-4.1, k=12025.12 | 73 | 90 | 66 | 74 | 10 | |
| SEAL-RAGModel=gpt-4.1-mini, k=12025.12 | 71 | 87 | 48 | 61 | 19 | |
| Self-RAGModel=gpt-4.1, k=12025.12 | 63 | 79 | 40 | 53 | — | |
| SEAL-RAGModel=gpt-4o-mini, k=12025.12 | 62 | 86 | 44 | 58 | 7 | |
| Self-RAGModel=gpt-4o, k=12025.12 | 59 | 75 | 37 | 50 | — | |
| CRAGModel=gpt-4o, k=12025.12 | 58 | 54 | 27 | 36 | — | |
| CRAGModel=gpt-4.1, k=12025.12 | 58 | 57 | 28 | 38 | — | |
| CRAGModel=gpt-4o-mini, k=12025.12 | 55 | 42 | 21 | 28 | — | |
| CRAGModel=gpt-4.1-mini, k=12025.12 | 52 | 57 | 29 | 38 | — | |
| Self-RAGModel=gpt-4.1-mini, k=12025.12 | 49 | 72 | 36 | 48 | — | |
| Self-RAGModel=gpt-4o-mini, k=12025.12 | 48 | 61 | 31 | 41 | — | |
| Basic RAGModel=gpt-4o-mini, k=12025.12 | 41 | 85 | 42 | 57 | — | |
| Basic RAGModel=gpt-4o, k=12025.12 | 41 | 85 | 42 | 57 | — | |
| Basic RAGModel=gpt-4.1, k=12025.12 | 40 | 85 | 42 | 57 | — | |
| Basic RAGModel=gpt-4.1-mini, k=12025.12 | 39 | 85 | 42 | 57 | — |