Tool Use on BrowseComp Domains (Domain-specific (9) + Full Search)
27.8AccuracyGold Oracle
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gold OracleModel=GPT-5, Strategy=ReAct2026.02 | 27.8 | 6.8 | |
| TOOLOBSERVERModel=GPT-5, Strategy=ReAct2026.02 | 22.1 | 9.5 | |
| Base ReActModel=GPT-5, Strategy=ReAct2026.02 | 21.8 | 9.3 | |
| EasyToolModel=GPT-5, Strategy=ReAct2026.02 | 17.1 | 8.6 | |
| Play2PromptModel=GPT-5, Strategy=ReAct2026.02 | 15.8 | 7.2 | |
| TOOLOBSERVERModel=GPT-5-mini, Strategy=ReAct2026.02 | 12.8 | 3.3 | |
| Gold OracleModel=GPT-5-mini, Strategy=ReAct2026.02 | 12.6 | 3 | |
| Base ReActModel=GPT-5-mini, Strategy=ReAct2026.02 | 9.7 | 3.7 | |
| EasyToolModel=GPT-5-mini, Strategy=ReAct2026.02 | 8.2 | 3.5 | |
| Play2PromptModel=GPT-5-mini, Strategy=ReAct2026.02 | 5.6 | 3.5 |