PCT Political Inconsistency (CAIS)
Safety · 2026-05-21
Polarized Contrastive Pairs evaluates covert political manipulation by comparing model responses to matched prompts about opposing political subjects. SAFE reports Political Inconsistency, the complement of Political Consistency across 1,000 paired evaluations, so lower scores indicate more symmetric sentiment and helpfulness.
Top models (lower is better)
| Model | Score |
|---|---|
| Muse Spark 1.1 | 27.9 |
| Grok 4 Fast | 30.8 |
| Grok 4.1 Fast | 32.5 |
| GPT-5 | 35.3 |
| Gemini 2.5 Pro | 37.3 |
| GPT-5.5 | 39.0 |
| o3 | 39.0 |
| Claude Fable 5 | 39.8 |
| Grok 4.20 | 39.9 |
| GPT-5.6 Sol | 41.7 |
| Gemini 3.1 Pro Preview | 43.3 |
| Kimi K2.6 | 43.6 |
| DeepSeek-V4-Pro | 44.0 |
| GPT-5.6 Luna | 44.7 |
| GPT-5.6 Terra | 46.2 |
| GPT-5.1 | 47.9 |
| Opus 4.7 | 48.2 |
| Opus 4.5 | 49.4 |
| Grok 4 | 49.5 |
| GLM-5.2 | 50.0 |
| Grok 4.5 | 50.2 |
| Grok 4.3 | 51.6 |
| Opus 4.6 | 51.7 |
| Gemini 3.1 Flash-Lite Preview | 51.9 |
| Kimi K3 | 52.4 |
| Gemini 3.5 Flash | 53.4 |
| Opus 4.8 | 53.8 |
| GPT-5.2 | 57.0 |
| GPT-5.4 | 59.4 |
| GPT-4o (2024-11-20) | 59.6 |
| GPT-5.4 Mini | 61.1 |
| Haiku 4.5 | 62.2 |
| GPT-5.4 Nano | 63.3 |