MuPPET - Utility: leaderboard
MuPPET (Multi-Party Privacy Exposure Testing, 2026) tests whether LLM assistants that hold memories of private user information leak it in multi-party workplace group conversations, judged by an LLM under contextual integrity. This board is the share (%) of answers that use at least one useful preference or constraint from the memories, with no defence prompt; higher is better.
Metric: Utility (%): share of answers that use at least one useful preference or constraint extracted from the memories, judged by an LLM, in the main multi-party setting with no defence prompt: an LLM assistant acting for one user in a MuPPET multi-party workplace group conversation holds memory summaries with private information about that user and answers the group; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 82.03 | #49 |
| 2 | Qwen 3 14B | 81.14 | #1233 |
| 3 | Qwen 3 8B | 78.29 | #1830 |
| 4 | Gemini 2.5 Pro | 74.91 | #266 |
| 5 | Qwen 3 4B | 74.38 | #2281 |
| 6 | Llama 3.1 8B Instruct | 69.93 | #4246 |
| 7 | Llama 3 8B Instruct | 67.26 | #4612 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=muppet-utility · How It Works · Data refreshed daily, snapshot 2026-11-13.