Pressure Reveals Character - Scheming: leaderboard
Mean 1-5 judge rubric score over the 166 Scheming scenarios of Pressure Reveals Character (scheming, deceptive alignment, power seeking, reward hacking and sabotage). A subset of the headline's transcripts.
Metric: Mean judge rubric score over 166 scenarios (1 = failure, 5 = exemplary). Source: storage.googleapis.com. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 4.8 | #156 |
| 2 | Claude Sonnet 4.5 | 4.73 | #250 |
| 3 | GPT-5.2 | 4.55 | #176 |
| 4 | Claude Sonnet 4 | 4.42 | #400 |
| 5 | GPT-5 Mini | 4.19 | #348 |
| 6 | Gemini 3 Pro | 4.11 | #137 |
| 7 | DeepSeek V3.2 | 4.01 | #406 |
| 8 | Kimi K2 | 3.9 | #439 |
| 9 | Qwen 3 235B A22B | 3.9 | #623 |
| 10 | GPT-OSS-120B | 3.87 | #700 |
| 11 | Gemini 2.5 Pro | 3.79 | #266 |
| 12 | Gemini 3 Flash | 3.64 | #164 |
| 13 | Grok 4.1 | 3.63 | #476 |
| 14 | Gemini 2.5 Flash | 3.61 | #453 |
| 15 | GPT-4o | 3.58 | #741 |
Interactive version: theaggregate.ai/benchmark?slug=pressure-reveals-character-scheming · How It Works · Data refreshed daily, snapshot 2026-11-13.