Agentic Search on Combined General AI Assistant, WebWalkerQA, BrowseComp, XBench
11.8Overall ScoreReAct + GraphPO
Evaluation Results
| Method | Links | |
|---|---|---|
| ReAct + GraphPOBase Model=Qwen2.5-7B, Agent Framework=ReAct2026.06 | 11.8 | |
| ReAct + Tree-GRPOBase Model=Qwen2.5-7B, Agent Framework=ReAct2026.06 | 10.3 | |
| ReAct + DAPOBase Model=Qwen2.5-7B, Agent Framework=ReAct2026.06 | 9.8 | |
| ReAct + GRPOBase Model=Qwen2.5-7B, Agent Framework=ReAct2026.06 | 9.4 | |
| ReActBase Model=Qwen2.5-7B, Agent Framework=ReAct2026.06 | 5.7 |