Pressure Reveals Character - Non-Manipulation: Epistemic Manipulation: leaderboard
Mean 1-5 judge rubric score over 28 multi-turn scenarios testing not manipulating the user's beliefs. One behaviour of the Non-Manipulation category; a subset of the headline's transcripts.
Metric: Mean judge rubric score over 28 scenarios (1 = failure, 5 = exemplary). Source: storage.googleapis.com. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4 | 5 | #400 |
| 2 | Claude Sonnet 4.5 | 4.93 | #250 |
| 3 | Claude Opus 4.5 | 4.89 | #156 |
| 4 | Gemini 3 Pro | 4.64 | #137 |
| 5 | GPT-5.2 | 4.61 | #176 |
| 6 | Gemini 3 Flash | 4.36 | #164 |
| 7 | Grok 4.1 | 4.36 | #476 |
| 8 | DeepSeek V3.2 | 4.32 | #406 |
| 9 | Qwen 3 235B A22B | 4.29 | #623 |
| 10 | Kimi K2 | 4.25 | #439 |
| 11 | Qwen 3 Next 80B A3B | 4.25 | #757 |
| 12 | GPT-5 Mini | 4.21 | #348 |
| 13 | Qwen 3 32B | 4.18 | #960 |
| 14 | Gemini 2.5 Pro | 4.11 | #266 |
| 15 | Gemini 2.5 Flash | 3.86 | #453 |
Interactive version: theaggregate.ai/benchmark?slug=pressure-reveals-character-non-manipulation-epistemic-manipulation · How It Works · Data refreshed daily, snapshot 2026-11-13.