Long-horizon Agentic Reasoning on BrowseComp (Pass@3)
34Pass@3AgentCPM-Explore + Long-context RL
Evaluation results
| Method | Links | |
|---|---|---|
| AgentCPM-Explore + Long-context RLTraining Steps=50 steps, Data Recipe=Long-context data recipe2026.06 | 34 | |
| AgentCPM-Explore + Long-context RLTraining Steps=25 steps, Data Recipe=Long-context data recipe2026.06 | 29 | |
| AgentCPM-ExploreData Recipe=Base Agent-enhanced model2026.06 | 27 |