Phare Prompt Injection (Spanish)
Safety · 2025-12-09
Phare jailbreak task testing resistance to prompt-injection attacks. This row is the Spanish language split.
Top models (higher is better)
| Model | Score |
|---|---|
| Sonnet 4.5 | 98.2 |
| Opus 4.1 | 97.9 |
| Haiku 4.5 | 97.6 |
| Opus 4.5 | 97.6 |
| Opus 4.6 | 96.1 |
| Sonnet 5 | 93.4 |
| Sonnet 4.6 | 92.7 |
| Haiku 3.5 | 91.9 |
| Qwen3.7 Max Preview | 90.6 |
| GPT-5.5 | 89.0 |
| GPT-5.2 | 87.3 |
| GPT-5 Mini | 87.2 |
| Kimi K2.6 | 86.5 |
| Claude 3.7 Sonnet | 86.1 |
| GPT-5 Nano | 84.9 |
| Qwen3.7-Plus | 84.9 |
| Llama 3.1 405B Instruct | 84.6 |
| GPT-5.1 | 84.0 |
| gpt-oss-120b | 84.0 |
| Kimi K2.5 | 82.8 |
| GPT-5 | 79.8 |
| Llama 4 Maverick | 79.8 |
| Grok 4.3 | 77.1 |
| Grok 4 | 74.4 |
| DeepSeek-V4-Flash | 72.6 |
| GLM-5.2 | 72.3 |
| DeepSeek-V4-Pro | 71.4 |
| Gemma 4 31B IT | 71.3 |
| GPT-4.1 Nano | 70.5 |
| GPT-4o | 68.7 |
| Gemini 3.1 Pro Preview | 67.5 |
| Gemini 3 Pro Preview | 66.6 |
| Qwen-Plus (2025-01-25) | 66.3 |
| Llama 4 Scout | 64.5 |
| GPT-4.1 | 63.9 |
| GPT-4o Mini | 63.9 |
| Qwen3 Max (rolling alias) | 61.5 |
| DeepSeek-R1-0528 | 60.2 |
| Gemini 3.1 Flash-Lite | 60.2 |
| GPT-4.1 Mini | 60.2 |