Phare Encoding Jailbreaks (English)
Safety · 2025-12-09
Phare jailbreak task using encoded or obfuscated prompts to bypass safeguards. This row is the English language split.
Top models (higher is better)
| Model | Score |
|---|---|
| Sonnet 5 | 75.4 |
| Magistral Small 1.2 | 70.0 |
| Magistral Medium 1.2 | 67.5 |
| Qwen3 8B | 66.3 |
| Opus 4.1 | 61.7 |
| Llama 3.1 8B Instruct | 61.3 |
| Llama 3.1 405B Instruct | 54.6 |
| GPT-4o | 53.3 |
| Haiku 4.5 | 52.9 |
| Opus 4.5 | 49.2 |
| Qwen3.7-Plus | 46.7 |
| Gemini 3 Pro Preview | 46.0 |
| Gemma 3 12B IT | 44.6 |
| Gemma 4 31B IT | 44.6 |
| Sonnet 4.5 | 42.1 |
| Mistral Small 3.2 24B Instruct 2506 | 40.4 |
| Grok 4 | 40.0 |
| Grok 3 Mini | 39.2 |
| Opus 4.6 | 37.9 |
| Qwen3.7 Max Preview | 37.8 |
| Llama 4 Scout | 37.5 |
| Kimi K2.6 | 35.8 |
| GPT-5 | 34.6 |
| Qwen3 VL 30B A3B Instruct | 34.6 |
| Gemma 3 27B IT | 34.2 |
| Mistral Medium 3.5 | 33.5 |
| GPT-5.2 | 33.3 |
| Llama 3.3 70B Instruct | 32.5 |
| Command A | 31.7 |
| GPT-4.1 Nano | 30.4 |
| gpt-oss-120b | 30.4 |
| Sonnet 4.6 | 30.0 |
| GPT-4o Mini | 28.3 |
| Mistral Medium 3.1 | 27.9 |
| Qwen2.5-Max | 27.6 |
| GPT-5 Nano | 26.3 |
| DeepSeek-R1-0528 | 25.8 |
| Mistral Large 2 (Instruct 2407) | 25.8 |
| DeepSeek-V4-Flash | 25.4 |
| GPT-4.1 | 24.2 |