Phare Self-Assessed Stereotypes (French)
Safety · 2025-03-27
Phare bias task testing resistance to self-assessed stereotypes and biased outputs. This row is the French language split.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-4.1 Mini | 87.0 |
| Grok 4 Fast | 79.6 |
| Llama 4 Maverick | 77.5 |
| Llama 3.1 405B Instruct | 77.1 |
| Haiku 4.5 | 75.0 |
| Mistral Small 3.2 24B Instruct 2506 | 73.3 |
| Opus 4.5 | 69.5 |
| Llama 4 Scout | 69.1 |
| Mistral Large 3 675B Instruct 2512 | 65.2 |
| DeepSeek-V3 | 64.0 |
| Qwen3 8B | 61.8 |
| Kimi K2.6 | 61.7 |
| Qwen-Plus (2025-01-25) | 60.3 |
| Mistral Medium 3.5 | 60.1 |
| DeepSeek-V3.1 | 60.0 |
| Gemini 2.5 Flash | 57.3 |
| Magistral Medium 1.2 | 56.8 |
| GLM-5.2 | 56.7 |
| GPT-4.1 | 55.5 |
| Qwen3.7-Plus | 54.7 |
| Qwen3.7 Max Preview | 53.4 |
| Llama 3.1 8B Instruct | 51.7 |
| Grok 3 Mini | 51.0 |
| Qwen2.5-Max | 50.4 |
| Magistral Small 1.2 | 50.1 |
| Qwen3 VL 30B A3B Instruct | 49.6 |
| Gemini 2.5 Flash-Lite | 49.5 |
| Gemini 3 Pro Preview | 49.3 |
| Gemini 3.1 Pro Preview | 47.7 |
| Command A | 47.7 |
| Gemini 2.0 Flash | 47.2 |
| GPT-4o | 46.0 |
| Grok 4.3 | 45.5 |
| Mistral Large 2 (Instruct 2407) | 45.5 |
| Llama 3.3 70B Instruct | 45.2 |
| Sonnet 4.5 | 44.1 |
| Opus 4.1 | 44.0 |
| Gemini 3.1 Flash-Lite | 43.2 |
| DeepSeek-V3-0324 | 41.4 |
| Mistral Medium 3.1 | 41.0 |