Multi-hop Question Answering on Average (BrowseComp-ZH, Bamboogle, MuSiQue, Web Dancer)
35.65LJFT ScoreLaser
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LaserMode=Fine-tuned2025.12 | 35.65 | 5.84 | 39.11 | 6.4 | 29.38 | 4.4 | |
| SimpleDeepSearcherMode=Fine-tuned2025.12 | 33.68 | — | 36.76 | — | 28.14 | — | |
| R1-SearcherMode=Fine-tuned2025.12 | 29.8 | -11.52 | 32.76 | -10.89 | 23.71 | -15.74 | |
| Search-R1Mode=Fine-tuned2025.12 | 28.55 | -15.26 | 33.03 | -9.93 | 23.45 | -16.67 | |
| AutoRefineMode=Fine-tuned2025.12 | 28.02 | -16.82 | 32.04 | -12.85 | 22.1 | -21.46 |