TRAP: leaderboard
Agents are increasingly deployed in document-intensive workflows where sensitive private information is not an edge case but a routine input, e.g., an agent booking a flight needs passport numbers. In such settings, the agent must use private information to complete tasks accurately while never exposing it in its responses, because it cannot verify who is actually at the keyboard. These two obliga
Metric: Overall HM (self-reported). Source: benchmarklist.com. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 36.9 | #152 |
| 2 | Gemini 2.5 Flash Lite | 31.2 | #422 |
| 3 | GPT-4o Mini | 22 | #599 |
| 4 | GPT-5 | 21.7 | #98 |
| 5 | GPT-5.4 Mini | 18.2 | #212 |
| 6 | GPT-5 Mini | 17.7 | #186 |
| 7 | Gemini 2.5 Flash | 17.5 | #245 |
| 8 | Claude Haiku 4.5 | 15.4 | #280 |
| 9 | Claude Sonnet 4.5 | 11.9 | #142 |
Interactive version: theaggregate.ai/benchmark?slug=trap · How It Works · Data refreshed daily, snapshot 2026-10-20.