Phare Factuality (English)
Safety · 2025-02-19
Phare task testing factual answer reliability and hallucination resistance. This row is the English language split.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.5 | 90.8 |
| Gemini 3.1 Pro Preview | 89.7 |
| Opus 4.6 | 88.6 |
| Gemini 3 Pro Preview | 88.6 |
| Kimi K2.5 | 87.5 |
| Gemini 3.5 Flash | 87.2 |
| Opus 4.5 | 86.5 |
| Kimi K2.6 | 86.5 |
| DeepSeek-V4-Pro | 86.1 |
| Sonnet 4.6 | 85.8 |
| GPT-5 | 85.8 |
| GPT-5.1 | 85.8 |
| GPT-5.2 | 85.4 |
| Claude 3.7 Sonnet | 85.0 |
| Gemini 3.1 Flash-Lite | 85.0 |
| Grok 4 | 84.7 |
| Opus 4.1 | 84.0 |
| Gemini 2.5 Pro | 84.0 |
| GLM-5.2 | 84.0 |
| Sonnet 4.5 | 83.6 |
| GPT-4.1 | 83.6 |
| Grok 4.3 | 83.3 |
| Claude 3.5 Sonnet (Oct 2024) | 83.2 |
| GPT-4o | 82.6 |
| Qwen3.7 Max Preview | 82.6 |
| Grok 3 | 81.5 |
| Sonnet 5 | 80.8 |
| DeepSeek-V4-Flash | 80.1 |
| DeepSeek-R1-0528 | 80.0 |
| Gemini 1.5 Pro | 79.4 |
| Gemini 2.5 Flash | 79.4 |
| Mistral Large 2 (Instruct 2407) | 79.4 |
| GPT-5 Mini | 78.3 |
| Grok 2 | 78.3 |
| DeepSeek-V3 | 77.9 |
| DeepSeek-V3-0324 | 77.9 |
| Gemini 2.0 Flash | 77.9 |
| Qwen3 Max (rolling alias) | 77.9 |
| Mistral Large 3 675B Instruct 2512 | 77.6 |
| Qwen2.5-Max | 77.6 |