Refusal Self-Prediction - Calibration Error: leaderboard
Expected calibration error of the stated confidence in a refusal self-prediction. Lower is better. Source: arxiv.org/abs/2604.00228.
Metric: Expected Calibration Error (0-1). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 0.02 | #148 |
| 2 | Claude Sonnet 4 | 0.05 | #207 |
| 3 | GPT-5.2 | 0.1 | #115 |
| 4 | Llama 3.1 405B | 0.22 | #476 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=refusal-self-prediction-calibration-error · How It Works · Data refreshed daily, snapshot 2026-11-05.