Pressure Reveals Character - Corrigibility: Goal Stability: leaderboard
Mean 1-5 judge rubric score over 26 multi-turn scenarios testing keeping stable objectives under pressure. One behaviour of the Corrigibility category; a subset of the headline's transcripts.
Metric: Mean judge rubric score over 26 scenarios (1 = failure, 5 = exemplary). Source: storage.googleapis.com. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 5 | #250 |
| 2 | Claude Opus 4.5 | 5 | #156 |
| 3 | Claude Sonnet 4 | 4.92 | #400 |
| 4 | GPT-5.2 | 4.85 | #176 |
| 5 | GPT-5 Mini | 4.65 | #348 |
| 6 | Kimi K2 | 4.65 | #439 |
| 7 | DeepSeek V3.2 | 4.54 | #406 |
| 8 | Grok 4.1 | 4.5 | #476 |
| 9 | Gemini 3 Pro | 4.42 | #137 |
| 10 | Llama 4 Maverick | 4.35 | #1099 |
| 11 | Qwen 3 235B A22B | 4.35 | #623 |
| 12 | Gemini 3 Flash | 4.31 | #164 |
| 13 | Gemini 2.5 Pro | 4.27 | #266 |
| 14 | GPT-4o | 4.23 | #741 |
| 15 | GPT-OSS-120B | 4.19 | #700 |
Interactive version: theaggregate.ai/benchmark?slug=pressure-reveals-character-corrigibility-goal-stability · How It Works · Data refreshed daily, snapshot 2026-11-13.