Pressure Reveals Character - Scheming: Self-Preferential Bias: leaderboard
Mean 1-5 judge rubric score over 34 multi-turn scenarios testing not favouring its own continuation or outputs. One behaviour of the Scheming category; a subset of the headline's transcripts.
Metric: Mean judge rubric score over 34 scenarios (1 = failure, 5 = exemplary). Source: storage.googleapis.com. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Mistral Medium 3.1 | 4.88 | #852 |
| 2 | Claude Sonnet 4.5 | 4.85 | #250 |
| 3 | Claude Opus 4.5 | 4.76 | #156 |
| 4 | GPT-5.2 | 4.74 | #176 |
| 5 | DeepSeek V3.2 | 4.74 | #406 |
| 6 | Gemini 3 Pro | 4.71 | #137 |
| 7 | Claude Sonnet 4 | 4.68 | #400 |
| 8 | Llama 4 Maverick | 4.65 | #1099 |
| 9 | Gemini 3 Flash | 4.62 | #164 |
| 10 | Kimi K2 | 4.62 | #439 |
| 11 | Qwen 3 235B A22B | 4.62 | #623 |
| 12 | Gemini 2.5 Pro | 4.59 | #266 |
| 13 | GPT-4o | 4.56 | #741 |
| 14 | Gemini 2.5 Flash | 4.56 | #453 |
| 15 | Qwen 3 32B | 4.56 | #960 |
Interactive version: theaggregate.ai/benchmark?slug=pressure-reveals-character-scheming-self-preferential-bias · How It Works · Data refreshed daily, snapshot 2026-11-13.