Phare Encoding Jailbreaks (Spanish)
Safety · 2025-12-09
Phare jailbreak task using encoded or obfuscated prompts to bypass safeguards. This row is the Spanish language split.
Top models (higher is better)
| Model | Score |
|---|---|
| Llama 3.1 405B Instruct | 77.0 |
| Llama 3.1 8B Instruct | 72.1 |
| Sonnet 5 | 70.6 |
| Magistral Small 1.2 | 60.4 |
| Qwen3 8B | 55.5 |
| Magistral Medium 1.2 | 53.2 |
| GPT-4o | 42.6 |
| Opus 4.1 | 40.0 |
| Llama 4 Scout | 39.3 |
| Qwen3.7 Max Preview | 37.3 |
| Gemma 3 12B IT | 36.6 |
| Llama 3.3 70B Instruct | 35.5 |
| Mistral Small 3.2 24B Instruct 2506 | 35.5 |
| Grok 4 | 35.1 |
| Gemini 3 Pro Preview | 34.9 |
| Qwen3.7-Plus | 34.5 |
| Qwen3 VL 30B A3B Instruct | 33.6 |
| Mistral Medium 3.5 | 32.1 |
| Command A | 31.3 |
| Gemma 4 31B IT | 30.6 |
| Haiku 4.5 | 29.4 |
| Opus 4.5 | 29.4 |
| Gemma 3 27B IT | 29.4 |
| GPT-4.1 Nano | 29.1 |
| Grok 3 Mini | 29.1 |
| Mistral Large 2 (Instruct 2407) | 28.3 |
| Mistral Medium 3.1 | 28.3 |
| GPT-4o Mini | 27.9 |
| Kimi K2.6 | 26.0 |
| DeepSeek-V4-Flash | 24.5 |
| gpt-oss-120b | 23.4 |
| GPT-5 | 22.6 |
| Opus 4.6 | 22.3 |
| DeepSeek-V4-Pro | 22.3 |
| Qwen2.5-Max | 22.3 |
| Grok 2 | 21.9 |
| Llama 4 Maverick | 21.7 |
| GPT-5 Nano | 21.5 |
| GPT-5.2 | 20.8 |
| Mistral Large 3 675B Instruct 2512 | 20.8 |