Agentic Person Search on Track 1 (Who)
100SR@1Oracle
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OracleBackbone=GPT-4o2026.04 | 100 | 100 | 100 | |
| LLM ToolCallBackbone=GPT-4o2026.04 | 81.1 | 82.3 | 90.8 | |
| LLM DirectBackbone=GPT-4o2026.04 | 73.3 | 73.3 | 93 | |
| Rule-basedBackbone=GPT-4o2026.04 | 32.2 | 48 | 66 |