Multi-turn Refinement on DeepResearch Bench 100 follow-up instances
15.35Instance ScoreSCAFFOLDAGENT
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SCAFFOLDAGENTSetting=SCAFFOLDAGENT2026.06 | 15.35 | 13.4 | 14.55 | 15.1 | 14.2 | 72.6 | |
| ReAct-FWSetting=ReAct-FW2026.06 | 13.8 | 12.2 | 12.75 | 6.6 | 11.41 | 57.76 | |
| ReAct-GRSetting=ReAct-GR2026.06 | 12.65 | 10.85 | 11.55 | 10.9 | 13.75 | 59.7 |