Pressure Reveals Character - Non-Manipulation: Commercial Optimisation: leaderboard
Mean 1-5 judge rubric score over 33 multi-turn scenarios testing not putting commercial interests above the user's. One behaviour of the Non-Manipulation category; a subset of the headline's transcripts.
Metric: Mean judge rubric score over 33 scenarios (1 = failure, 5 = exemplary). Source: storage.googleapis.com. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 4.64 | #156 |
| 2 | Claude Sonnet 4.5 | 4.36 | #250 |
| 3 | Claude Sonnet 4 | 4.3 | #400 |
| 4 | GPT-5.2 | 3.82 | #176 |
| 5 | GPT-5 Mini | 3.7 | #348 |
| 6 | GPT-OSS-120B | 2.85 | #700 |
| 7 | Kimi K2 | 2.76 | #439 |
| 8 | Qwen 3 14B | 2.73 | #1233 |
| 9 | Qwen 3 32B | 2.55 | #960 |
| 10 | Qwen 3 235B A22B | 2.55 | #623 |
| 11 | Mistral Small 3.2 | 2.55 | #1591 |
| 12 | GPT-4o | 2.48 | #741 |
| 13 | Gemini 2.5 Flash | 2.48 | #453 |
| 14 | Llama 3.3 70B | 2.45 | #1498 |
| 15 | Llama 4 Maverick | 2.36 | #1099 |
Interactive version: theaggregate.ai/benchmark?slug=pressure-reveals-character-non-manipulation-commercial-optimisation · How It Works · Data refreshed daily, snapshot 2026-11-13.