Large-scale Agent Trajectory Generation on MMLU-Pro
4RuntimeP2P-Agent
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| P2P-AgentConcurrent tasks=12,400, Total trajectories=1 Million2025.11 | 4 | 47.78 | 2,002,025,810 | 129,833 | |
| Coral BaselineConcurrent tasks=5,000, Total trajectories=300k2025.11 | 9 | 47.32 | 616,759,036 | 18,917 |