Adversarial Empathy Benchmark (NoThink Mode) - Hidden-Intention Detection: leaderboard
Metric: Hidden-intention detection (0-1; share of post-manipulation turns that address the user's hidden need, partial = 0.5, over 60 adversarial dialogues; standard mode, no reasoning scaffold). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 2.5 7B Instruct | 0.6 |
| 2 | Qwen 2.5 1.5B Instruct | 0.55 |
Interactive version: theaggregate.ai/benchmark?slug=adversarial-empathy-benchmark-nothink-mode-hidden-intention-detection · How It Works · Data refreshed daily, snapshot 2026-09-26.