Phare Encoding Jailbreaks (French)
Safety · 2025-12-09
Phare jailbreak task using encoded or obfuscated prompts to bypass safeguards. This row is the French language split.
Top models (higher is better)
| Model | Score |
|---|---|
| Sonnet 5 | 67.3 |
| Llama 3.1 8B Instruct | 64.5 |
| Magistral Small 1.2 | 62.7 |
| Qwen3 8B | 62.4 |
| Llama 3.1 405B Instruct | 57.4 |
| Opus 4.1 | 53.6 |
| Magistral Medium 1.2 | 53.6 |
| Opus 4.5 | 52.1 |
| Qwen3.7 Max Preview | 50.6 |
| Qwen3.7-Plus | 46.2 |
| Gemini 3 Pro Preview | 45.4 |
| Haiku 4.5 | 43.4 |
| Gemma 3 12B IT | 42.2 |
| Opus 4.6 | 39.5 |
| Gemma 4 31B IT | 39.5 |
| Kimi K2.6 | 38.0 |
| GPT-4o | 37.3 |
| Mistral Small 3.2 24B Instruct 2506 | 36.5 |
| Grok 4 | 35.4 |
| Llama 4 Scout | 35.4 |
| Mistral Medium 3.5 | 34.6 |
| Sonnet 4.5 | 33.8 |
| GPT-4o Mini | 33.2 |
| Llama 3.3 70B Instruct | 33.2 |
| Command A | 33.1 |
| Qwen2.5-Max | 33.1 |
| Qwen3 VL 30B A3B Instruct | 32.7 |
| GPT-4.1 Nano | 32.4 |
| Gemma 3 27B IT | 32.3 |
| GPT-5 | 31.2 |
| Grok 3 Mini | 31.2 |
| GPT-5.2 | 29.3 |
| Mistral Large 2 (Instruct 2407) | 28.5 |
| Mistral Medium 3.1 | 28.5 |
| DeepSeek-R1-0528 | 26.6 |
| DeepSeek-V3.1 | 26.2 |
| Grok 2 | 26.2 |
| DeepSeek-V4-Flash | 25.9 |
| Llama 4 Maverick | 25.6 |
| gpt-oss-120b | 25.1 |