Pressure Reveals Character - Robustness: leaderboard
Mean 1-5 judge rubric score over the 75 Robustness scenarios of Pressure Reveals Character (instruction hierarchy, prefill attacks and adversarial robustness). A subset of the headline's transcripts.
Metric: Mean judge rubric score over 75 scenarios (1 = failure, 5 = exemplary). Source: storage.googleapis.com. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 4.6 | #348 |
| 2 | GPT-4o | 4.56 | #741 |
| 3 | GPT-OSS-120B | 4.35 | #700 |
| 4 | GPT-5.2 | 4.23 | #176 |
| 5 | Claude Sonnet 4.5 | 4.03 | #250 |
| 6 | Grok 4.1 | 4.03 | #476 |
| 7 | Claude Opus 4.5 | 3.91 | #156 |
| 8 | Qwen 3 235B A22B | 3.88 | #623 |
| 9 | Llama 4 Maverick | 3.87 | #1099 |
| 10 | Claude Sonnet 4 | 3.81 | #400 |
| 11 | Gemini 3 Pro | 3.61 | #137 |
| 12 | DeepSeek V3.2 | 3.45 | #406 |
| 13 | Qwen 3 Next 80B A3B | 3.43 | #757 |
| 14 | Gemini 3 Flash | 3.37 | #164 |
| 15 | Llama 3.1 405B Instruct | 3.36 | #1115 |
Interactive version: theaggregate.ai/benchmark?slug=pressure-reveals-character-robustness · How It Works · Data refreshed daily, snapshot 2026-11-13.