Phare Prompt Injection (English)
Safety · 2025-12-09
Phare jailbreak task testing resistance to prompt-injection attacks. This row is the English language split.
Top models (higher is better)
| Model | Score |
|---|---|
| Haiku 4.5 | 97.9 |
| Opus 4.1 | 97.9 |
| Opus 4.5 | 97.3 |
| Sonnet 4.5 | 97.3 |
| Opus 4.6 | 97.3 |
| Sonnet 5 | 96.8 |
| Haiku 3.5 | 93.6 |
| GPT-5.5 | 93.5 |
| gpt-oss-120b | 92.5 |
| GPT-5.2 | 92.4 |
| Sonnet 4.6 | 91.9 |
| GPT-5 Nano | 90.4 |
| Kimi K2.6 | 90.4 |
| GPT-5 Mini | 90.2 |
| GPT-5.1 | 89.8 |
| Kimi K2.5 | 89.8 |
| Qwen3.7 Max Preview | 88.1 |
| Claude 3.7 Sonnet | 87.7 |
| Qwen3.7-Plus | 87.6 |
| GPT-5 | 86.0 |
| GLM-5.2 | 83.0 |
| GPT-4.1 Nano | 82.9 |
| GPT-4.1 | 79.7 |
| Qwen-Plus (2025-01-25) | 79.2 |
| Grok 4 | 79.1 |
| Llama 3.1 405B Instruct | 78.6 |
| Grok 4.3 | 78.1 |
| Qwen2.5-Max | 77.5 |
| Gemini 3 Pro Preview | 77.0 |
| Gemma 4 31B IT | 76.5 |
| Gemini 3.1 Pro Preview | 75.9 |
| GPT-4o | 75.4 |
| GPT-4o Mini | 73.3 |
| Qwen3 Max (rolling alias) | 72.5 |
| Gemini 3.1 Flash-Lite | 70.0 |
| GPT-4.1 Mini | 69.3 |
| Qwen3 VL 30B A3B Instruct | 63.6 |
| Grok 4 Fast | 63.1 |
| Gemini 3.5 Flash | 62.0 |
| Llama 4 Maverick | 61.5 |