Phare Framing Jailbreaks (Spanish)
Safety · 2025-12-09
Phare jailbreak task using framing strategies to elicit unsafe behavior. This row is the Spanish language split.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5 Nano | 100.0 |
| Opus 4.5 | 99.7 |
| Opus 4.6 | 99.7 |
| GPT-5.5 | 99.6 |
| GPT-5 Mini | 99.3 |
| Sonnet 5 | 98.9 |
| Haiku 4.5 | 98.6 |
| GPT-5.2 | 98.6 |
| GPT-5.1 | 98.2 |
| Sonnet 4.5 | 97.5 |
| GPT-5 | 96.1 |
| Llama 3.1 405B Instruct | 95.7 |
| Opus 4.1 | 94.0 |
| Qwen3.7 Max Preview | 92.5 |
| Sonnet 4.6 | 91.8 |
| Qwen3.7-Plus | 90.0 |
| Kimi K2.6 | 89.7 |
| Claude 3.7 Sonnet | 88.3 |
| Gemini 3 Pro Preview | 88.3 |
| Haiku 3.5 | 87.9 |
| gpt-oss-120b | 87.6 |
| Grok 4.3 | 80.8 |
| Grok 4 | 78.4 |
| GLM-5.2 | 76.4 |
| Gemini 3.1 Pro Preview | 73.8 |
| Gemma 4 31B IT | 72.3 |
| Kimi K2.5 | 70.8 |
| Llama 3.3 70B Instruct | 69.2 |
| GPT-4o | 66.0 |
| Gemini 3.5 Flash | 65.5 |
| Llama 4 Maverick | 63.5 |
| Llama 3.1 8B Instruct | 62.6 |
| Gemini 3.1 Flash-Lite | 61.7 |
| Qwen3 Max (rolling alias) | 58.5 |
| Qwen-Plus (2025-01-25) | 55.7 |
| GPT-4o Mini | 55.3 |
| Gemini 2.5 Pro | 51.4 |
| Llama 4 Scout | 50.0 |
| Gemini 2.5 Flash-Lite | 49.6 |
| Gemini 2.5 Flash | 47.2 |