SLM Trustworthiness - Privacy: leaderboard
SLM Trustworthiness (Benchmarking Trustworthiness of SLMs, arXiv 2608.11981v1, Tables I, II and IV): TrustLLM evaluation of uncompressed instruction-tuned models from 360M to 8B parameters, the baselines of a pruning, quantization and distillation study. Privacy: mean score on agreement tests about private information usage and on privacy scenario tasks, percent. Higher is better.
Metric: TrustLLM privacy accuracy (%; sub-task mean). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.2 1B Instruct | 60.21 | #1518 |
| 2 | Llama 3.2 3B Instruct | 60.15 | #1327 |
| 3 | Gemma 1.1 7B (IT) | 60.13 | #1348 |
| 4 | Llama 3.1 8B Instruct | 49.16 | #1017 |
| 5 | SmolLM2-360M-Instruct | 48.59 | #1594 |
| 6 | Qwen 2.5 7B Instruct | 41.53 | #863 |
| 7 | Qwen 2.5 1.5B Instruct | 40.89 | #1393 |
| 8 | Qwen 2.5 3B Instruct | 40.48 | #1159 |
| 9 | h2o-danube3-500m-chat | 29.02 | #1574 |
| 10 | Qwen 2.5 0.5B Instruct | 27.56 | #1510 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=slm-trustworthiness-privacy · How It Works · Data refreshed daily, snapshot 2026-11-05.