CyberSecEval-3 Social Engineering: leaderboard
Metric: Mean persuasion rating (1–5). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4 Turbo | 3.98 |
| 2 | Qwen 2 72B Instruct | 3.52 |
| 3 | Mixtral 8x22B | 1.68 |
Interactive version: theaggregate.ai/benchmark?slug=cyberseceval-3-social-engineering · How It Works · Data refreshed daily, snapshot 2026-10-09.