PrivacyPeek: leaderboard
LLM-based agents are rapidly advancing, autonomously invoking external tools to complete multi-step tasks for users. However, agents often acquire more sensitive information than the task requires. Existing privacy benchmarks audit what the agent's response or outgoing actions disclose, but overlook the acquisition stage where data first enters the agent's context. The over-acquired information is
Metric: TCR (self-reported). Source: benchmarklist.com. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4 | 86.46 | #199 |
| 2 | GPT-5.1 | 85.7 | #135 |
| 3 | GPT-4.1 | 83.76 | #248 |
| 4 | Claude Sonnet 4.5 | 80.03 | #142 |
| 5 | Qwen 3 30B A3B | 79.95 | #493 |
| 6 | Llama 3.3 70B Instruct | 79.61 | #534 |
| 7 | Qwen 3 14B | 79.53 | #537 |
| 8 | Llama 3.1 8B Instruct | 72.76 | #1035 |
| 9 | Qwen 3 4B Instruct | 71.32 | #657 |
| 10 | Llama 3.2 3B Instruct | 54.48 | #1352 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=privacypeek · How It Works · Data refreshed daily, snapshot 2026-10-20.