MASK
Safety · 2025-03-05
Honesty score on prompts that pressure models to contradict their beliefs.
Top models (higher is better)
| Model | Score |
|---|---|
| Opus 4.6 | 96.3 |
| Muse Spark 1.1 | 96.2 |
| Sonnet 4.5 | 96.1 |
| Haiku 4.5 | 95.4 |
| GPT-5.6 Sol | 95.4 |
| Sonnet 4 | 95.3 |
| Opus 4.1 | 94.2 |
| Opus 4.5 | 94.1 |
| GPT-5.6 Terra | 93.3 |
| Claude Fable 5 | 92.8 |
| GPT-5.6 Luna | 92.7 |
| gpt-oss-120b | 92.0 |
| Opus 4.8 | 91.9 |
| GPT-5.4 Pro | 91.7 |
| Kimi K3 | 90.6 |
| GPT-5.4 | 90.3 |
| GPT-5.5 | 90.1 |
| Sonnet 4.6 | 88.3 |
| GPT-5.1 | 88.0 |
| Opus 4 | 87.9 |
| GPT-5 Nano | 87.7 |
| GPT-5.2 | 87.1 |
| gpt-oss-20b | 86.5 |
| GPT-5 Pro | 86.0 |
| o3 | 84.5 |
| GLM-5.1 | 84.2 |
| GPT-5.4 Mini | 84.1 |
| Claude 3.7 Sonnet | 83.1 |
| Opus 4.7 | 82.9 |
| GPT-5 Mini | 82.6 |
| o3-pro | 82.5 |
| Kimi K2.5 | 81.5 |
| GPT-5 | 79.3 |
| Opus 3 | 79.0 |
| O4 Mini | 78.6 |
| Kimi K2.6 | 76.8 |
| GPT-5.4 Nano | 76.3 |
| GLM-5.2 | 74.2 |
| Claude 3.5 Sonnet (Oct 2024) | 72.3 |
| Grok 4.5 | 72.1 |