Pressure Reveals Character - Non-Manipulation: Boundary Erosion: leaderboard
Mean 1-5 judge rubric score over 25 multi-turn scenarios testing resisting gradual expansion of inappropriate requests. One behaviour of the Non-Manipulation category; a subset of the headline's transcripts.
Metric: Mean judge rubric score over 25 scenarios (1 = failure, 5 = exemplary). Source: storage.googleapis.com. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 5 | #250 |
| 2 | Claude Opus 4.5 | 5 | #156 |
| 3 | GPT-5.2 | 4.92 | #176 |
| 4 | Claude Sonnet 4 | 4.84 | #400 |
| 5 | DeepSeek V3.2 | 4.52 | #406 |
| 6 | GPT-5 Mini | 4.4 | #348 |
| 7 | Gemini 3 Pro | 4.4 | #137 |
| 8 | GPT-OSS-120B | 4.16 | #700 |
| 9 | Gemini 2.5 Pro | 3.88 | #266 |
| 10 | Grok 4.1 | 3.84 | #476 |
| 11 | Kimi K2 | 3.76 | #439 |
| 12 | Qwen 3 235B A22B | 3.72 | #623 |
| 13 | Gemini 2.5 Flash | 3.6 | #453 |
| 14 | Llama 3.3 70B | 3.52 | #1498 |
| 15 | Gemini 3 Flash | 3.4 | #164 |
Interactive version: theaggregate.ai/benchmark?slug=pressure-reveals-character-non-manipulation-boundary-erosion · How It Works · Data refreshed daily, snapshot 2026-11-13.