SynthPAI: leaderboard
Synthetic Personal Attribute Inference benchmark measuring how accurately models infer private attributes from text, used to study privacy leakage.
Metric: Average accuracy in %. Source: huggingface.co. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 | 77.9 | #438 |
| 2 | Llama 3 70B | 72.2 | #631 |
| 3 | Mixtral 8x22B | 72 | #766 |
| 4 | Claude 3 Opus | 71.1 | #571 |
| 5 | Claude 3 Sonnet | 70.9 | #775 |
| 6 | Gemini 1.5 Pro | 67.5 | #445 |
| 7 | Qwen 1.5 110B | 65.7 | #734 |
| 8 | Gemini 1.0 Pro | 64.6 | #933 |
| 9 | Claude 3 Haiku | 64 | #803 |
| 10 | Yi Chat 34B | 57.7 | #976 |
| 11 | Llama 2 70B | 56.8 | #1201 |
| 12 | GPT-3.5 | 55.9 | #818 |
| 13 | Llama 3 8B | 53.5 | #1242 |
| 14 | Mixtral 8x7B | 52.3 | #1143 |
| 15 | Llama 2 13B | 48.7 | #1436 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=synthpai · How It Works · Data refreshed daily, snapshot 2026-10-20.