Pressure Reveals Character - Corrigibility: leaderboard
Mean 1-5 judge rubric score over the 103 Corrigibility scenarios of Pressure Reveals Character (accepting corrections, appropriate autonomy, goal stability and self-preservation). A subset of the headline's transcripts.
Metric: Mean judge rubric score over 103 scenarios (1 = failure, 5 = exemplary). Source: storage.googleapis.com. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 4.64 | #176 |
| 2 | Claude Opus 4.5 | 4.56 | #156 |
| 3 | GPT-5 Mini | 4.5 | #348 |
| 4 | Kimi K2 | 4.5 | #439 |
| 5 | Claude Sonnet 4.5 | 4.49 | #250 |
| 6 | DeepSeek V3.2 | 4.48 | #406 |
| 7 | Claude Sonnet 4 | 4.4 | #400 |
| 8 | Qwen 3 235B A22B | 4.35 | #623 |
| 9 | Gemini 2.5 Pro | 4.34 | #266 |
| 10 | Gemini 3 Flash | 4.32 | #164 |
| 11 | Gemini 3 Pro | 4.32 | #137 |
| 12 | GPT-4o | 4.24 | #741 |
| 13 | Gemini 2.5 Flash | 4.2 | #453 |
| 14 | Qwen 3 14B | 4.2 | #1233 |
| 15 | Llama 4 Maverick | 4.17 | #1099 |
Interactive version: theaggregate.ai/benchmark?slug=pressure-reveals-character-corrigibility · How It Works · Data refreshed daily, snapshot 2026-11-13.