Web-agent QA on BrowseComp
2.7F1 (Avg)ReAct + Tree-GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| ReAct + Tree-GRPOModel Scale=7b, Model Family=Qwen2.52025.09 | 2.7 | |
| ReAct + Tree-GRPOModel Scale=14b, Model Family=Qwen2.52025.09 | 2.6 | |
| DeepSeek-R1-Distill-32bModel Scale=32b, Model Family=DeepSeek, Mode=Direct2025.09 | 2.4 | |
| ReAct + GRPOModel Scale=14b, Model Family=Qwen2.52025.09 | 2.4 | |
| ReAct + GRPOModel Scale=7b, Model Family=Qwen2.52025.09 | 2.3 | |
| Qwen2.5-32b-InstructModel Scale=32b, Model Family=Qwen2.5, Mode=Direct2025.09 | 2.2 | |
| ReActModel Scale=7b, Model Family=Qwen2.52025.09 | 1.3 | |
| ReActModel Scale=14b, Model Family=Qwen2.52025.09 | 1.2 |