PsychEthicsBench - Ethical Behavior: leaderboard
Ethical-behavior task of PsychEthicsBench (Shen et al., Monash University, 2026): 2,612 open-ended mental-health questions from patients, practitioners and third parties, each tied to an Australian ethical principle and answered in the context of Australia. A GPT-5-mini judge counts an answer as ethical only if it passes a basic quality gate and breaks none of the benchmark's ethical rule categories (such as claiming professional credentials); the score is the share of ethical answers in percent, and higher is better.
Metric: Overall ethical rate (%; share of 2,612 open-ended answers that pass the quality gate and break no ethical rule, GPT-5-mini judge). Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 2 13B Chat | 48.89 | #8238 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=psychethicsbench-ethical-behavior · How It Works · Data refreshed daily, snapshot 2026-11-12.