Multi-document Multi-entity Question Answering on Loong 100K-200K Tokens
81.44Spotlight Locating Avg ScoreDocSage
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| DocSageMethod=DocSage (Ours)2026.03 | 81.44 | 80 | 68.12 | 30 | 61.72 | 41 | 68.15 | 52 | 62.43 | 48 | |
| GPT-4oMethod=GPT-4o2026.03 | 76.34 | 66 | 43.25 | 21 | 39.47 | 4 | 45.96 | 9 | 47.89 | 19 | |
| GPT-4o + RAGMethod=GPT-4o + RAG2026.03 | 75.16 | 56 | 43.04 | 28 | 33.44 | 2 | 39.53 | 14 | 44.73 | 18 | |
| StructRAGMethod=StructRAG2026.03 | 67.25 | 43 | 56.59 | 34 | 57.1 | 10 | 48.74 | 13 | 56.76 | 21 | |
| GraphRAGMethod=GraphRAG2026.03 | 16.15 | 0 | 27.95 | 0 | 43.35 | 0 | 44.2 | 17 | 33.95 | 4 |