Phare Misinformation
Safety · 2025-02-19
Phare task testing resistance to misinformation generation or endorsement.
Top models (higher is better)
| Model | Score |
|---|---|
| Haiku 4.5 | 95.5 |
| Claude 3.7 Sonnet | 89.5 |
| Sonnet 4.5 | 86.2 |
| Opus 4.5 | 85.8 |
| Claude 3.5 Sonnet (Oct 2024) | 85.4 |
| Opus 4.1 | 85.0 |
| Sonnet 5 | 84.4 |
| GPT-5 Nano | 80.0 |
| Opus 4.6 | 79.6 |
| Qwen-Plus (2025-01-25) | 77.9 |
| GPT-5 Mini | 77.5 |
| Magistral Medium 1.2 | 76.7 |
| GPT-4.1 Nano | 76.6 |
| Haiku 3.5 | 76.4 |
| Gemini 3.1 Pro Preview | 76.0 |
| Kimi K2.6 | 74.9 |
| Gemini 1.5 Pro | 74.4 |
| Llama 3.1 8B Instruct | 74.2 |
| Llama 3.1 405B Instruct | 73.6 |
| Sonnet 4.6 | 73.2 |
| Gemini 2.5 Flash-Lite | 72.2 |
| DeepSeek-V4-Flash | 69.9 |
| Gemma 4 31B IT | 68.8 |
| Mistral Large 2 (Instruct 2407) | 68.7 |
| Gemini 2.5 Flash | 68.4 |
| GPT-4o | 68.1 |
| Command A | 67.7 |
| GPT-5.1 | 67.7 |
| Gemini 3.1 Flash-Lite | 67.0 |
| Gemini 2.5 Pro | 65.2 |
| Llama 4 Maverick | 65.0 |
| Grok 4 | 64.8 |
| Qwen3 Max (rolling alias) | 64.4 |
| Llama 3.3 70B Instruct | 64.2 |
| Grok 4.3 | 63.7 |
| Qwen3.7 Max Preview | 63.5 |
| Grok 3 Mini | 63.2 |
| Gemini 2.0 Flash | 62.4 |
| Gemini 3 Pro Preview | 62.1 |
| Kimi K2.5 | 61.7 |