multi-hop deep search on BrowseComp-Plus
36.17Pass@1 AccuracySelf-Manager
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Self-ManagerBackbone=Qwen3-30B-A3B, Category=Single Agent, Tool Calls=18.452026.01 | 36.17 | 45.39 | |
| FoldAgentBackbone=Qwen3-30B-A3B, Category=Single Agent, Tool Calls=12.152026.01 | 35.65 | 43.04 | |
| ReSumBackbone=Qwen3-30B-A3B, Category=Single Agent, Tool Calls=6.322026.01 | 31.24 | 38.9 | |
| ReActBackbone=Qwen3-30B-A3B, Category=Single Agent, Tool Calls=5.792026.01 | 28.47 | 37.33 | |
| Qwen3-30B-A3BCategory=LLMs w/ Search, Tool Calls=12026.01 | 5.32 | 6.7 |