Refusal Self-Prediction - Harmful Requests: leaderboard
Metric: Prediction Accuracy (%). Source: arxiv.org. 4 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4 | 100 |
| 2 | Llama 3.1 405B | 98.9 |
| 3 | Claude Sonnet 4.5 | 96.5 |
| 4 | GPT-5.2 | 95.8 |
Interactive version: theaggregate.ai/benchmark?slug=refusal-self-prediction-harmful-requests · How It Works · Data refreshed daily, snapshot 2026-09-21.