Phare Self-Assessed Stereotypes (Spanish)
Safety · 2025-03-27
Phare bias task testing resistance to self-assessed stereotypes and biased outputs. This row is the Spanish language split.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-4.1 Mini | 88.3 |
| Llama 3.1 405B Instruct | 81.8 |
| Grok 4 Fast | 79.6 |
| Mistral Small 3.2 24B Instruct 2506 | 75.2 |
| DeepSeek-V3.1 | 75.1 |
| Llama 4 Maverick | 74.6 |
| Haiku 4.5 | 74.2 |
| DeepSeek-V3-0324 | 68.7 |
| Llama 4 Scout | 63.2 |
| Qwen3.7-Plus | 61.4 |
| Magistral Medium 1.2 | 61.0 |
| Qwen3 VL 30B A3B Instruct | 60.7 |
| Mistral Large 3 675B Instruct 2512 | 59.6 |
| Mistral Medium 3.5 | 59.6 |
| Opus 4.5 | 58.7 |
| Qwen-Plus (2025-01-25) | 58.1 |
| Gemini 2.0 Flash | 56.4 |
| Kimi K2.6 | 55.2 |
| DeepSeek-V3 | 54.4 |
| GPT-5 Mini | 54.2 |
| Qwen3 8B | 53.2 |
| Magistral Small 1.2 | 51.8 |
| GPT-4o | 51.6 |
| Gemini 2.5 Flash-Lite | 51.1 |
| GPT-5.1 | 50.7 |
| Llama 3.1 8B Instruct | 50.4 |
| Gemini 3 Pro Preview | 50.2 |
| Gemini 2.5 Flash | 50.1 |
| GPT-4.1 | 49.7 |
| Qwen3 Max (rolling alias) | 49.1 |
| Grok 4.3 | 48.1 |
| Llama 3.3 70B Instruct | 48.1 |
| Qwen2.5-Max | 48.0 |
| Gemini 3.1 Pro Preview | 46.1 |
| Sonnet 4.5 | 45.7 |
| Qwen3.7 Max Preview | 43.8 |
| GPT-5.2 | 43.6 |
| GLM-5.2 | 43.2 |
| Gemma 3 27B IT | 43.1 |
| gpt-oss-120b | 42.8 |