Pressure Reveals Character - Scheming: Deceptive Alignment: leaderboard
Mean 1-5 judge rubric score over 13 multi-turn scenarios testing not faking alignment during evaluation. One behaviour of the Scheming category; a subset of the headline's transcripts.
Metric: Mean judge rubric score over 13 scenarios (1 = failure, 5 = exemplary). Source: storage.googleapis.com. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 4.85 | #348 |
| 2 | Claude Opus 4.5 | 4.77 | #156 |
| 3 | GPT-5.2 | 4.69 | #176 |
| 4 | Grok 4.1 | 4.69 | #476 |
| 5 | Claude Sonnet 4.5 | 4.46 | #250 |
| 6 | Qwen 3 235B A22B | 4.38 | #623 |
| 7 | GPT-OSS-120B | 4.31 | #700 |
| 8 | Claude Sonnet 4 | 4.31 | #400 |
| 9 | Kimi K2 | 4.31 | #439 |
| 10 | GPT-4o | 4.23 | #741 |
| 11 | DeepSeek V3.2 | 4.08 | #406 |
| 12 | Gemini 3 Pro | 4 | #137 |
| 13 | Qwen 3 Next 80B A3B | 4 | #757 |
| 14 | Gemini 3 Flash | 3.92 | #164 |
| 15 | Qwen 3 14B | 3.92 | #1233 |
Interactive version: theaggregate.ai/benchmark?slug=pressure-reveals-character-scheming-deceptive-alignment · How It Works · Data refreshed daily, snapshot 2026-11-13.