Phare Jailbreak Resistance
Safety · 2025-12-16
Phare module score for jailbreak resistance. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| Sonnet 5 | 86.8 |
| Opus 4.1 | 81.3 |
| Opus 4.5 | 79.8 |
| Haiku 4.5 | 78.5 |
| Llama 3.1 405B Instruct | 76.5 |
| Opus 4.6 | 75.2 |
| Sonnet 4.5 | 75.2 |
| Qwen3.7 Max Preview | 73.2 |
| GPT-5.2 | 71.6 |
| Kimi K2.6 | 71.4 |
| Qwen3.7-Plus | 70.6 |
| GPT-5 Nano | 69.6 |
| Sonnet 4.6 | 69.4 |
| GPT-5 | 68.7 |
| GPT-5 Mini | 68.6 |
| gpt-oss-120b | 65.9 |
| GPT-5.1 | 65.8 |
| GPT-5.5 | 65.1 |
| Gemini 3 Pro Preview | 65.1 |
| Grok 4 | 64.8 |
| Haiku 3.5 | 64.8 |
| Claude 3.7 Sonnet | 63.5 |
| Kimi K2.5 | 61.3 |
| Gemma 4 31B IT | 61.2 |
| GPT-4o | 60.5 |
| Llama 3.1 8B Instruct | 58.8 |
| GLM-5.2 | 56.5 |
| Grok 4.3 | 54.5 |
| GPT-4.1 Nano | 53.3 |
| GPT-4o Mini | 52.1 |
| Gemini 3.1 Pro Preview | 51.6 |
| Qwen-Plus (2025-01-25) | 50.0 |
| Qwen3 8B | 49.1 |
| Llama 4 Maverick | 49.0 |
| Llama 4 Scout | 48.9 |
| Qwen3 Max (rolling alias) | 48.4 |
| Gemini 3.1 Flash-Lite | 47.7 |
| Llama 3.3 70B Instruct | 47.7 |
| Qwen2.5-Max | 46.3 |
| GPT-4.1 | 46.1 |