Phare Hallucination Resistance
Safety · 2025-02-19
Phare module score for hallucination resistance. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| Opus 4.5 | 88.2 |
| Opus 4.6 | 87.8 |
| Gemini 3.1 Pro Preview | 87.6 |
| Sonnet 4.5 | 87.0 |
| Claude 3.5 Sonnet (Oct 2024) | 86.7 |
| Opus 4.1 | 86.2 |
| Sonnet 5 | 86.1 |
| Sonnet 4.6 | 85.5 |
| Claude 3.7 Sonnet | 85.2 |
| Haiku 4.5 | 83.6 |
| Kimi K2.6 | 82.8 |
| Gemini 3.5 Flash | 82.2 |
| DeepSeek-V4-Flash | 81.9 |
| GPT-5.1 | 81.8 |
| Qwen3.7 Max Preview | 81.1 |
| Gemini 3 Pro Preview | 81.0 |
| Grok 4 | 80.7 |
| Kimi K2.5 | 80.7 |
| Gemini 3.1 Flash-Lite | 80.6 |
| GPT-5.5 | 79.8 |
| DeepSeek-V4-Pro | 79.6 |
| GPT-5 Mini | 79.6 |
| Grok 4.3 | 79.5 |
| GPT-4o | 78.5 |
| GLM-5.2 | 78.2 |
| Gemini 1.5 Pro | 78.1 |
| Haiku 3.5 | 78.0 |
| GPT-4.1 | 77.6 |
| GPT-5.2 | 77.1 |
| Gemma 4 31B IT | 76.6 |
| GPT-5 Nano | 76.4 |
| Qwen3.7-Plus | 76.3 |
| Mistral Large 2 (Instruct 2407) | 75.6 |
| Qwen-Plus (2025-01-25) | 75.6 |
| Qwen3 Max (rolling alias) | 75.2 |
| Grok 3 | 75.1 |
| Gemini 2.5 Pro | 74.8 |
| GPT-5 | 74.6 |
| DeepSeek-R1-0528 | 72.9 |
| Gemini 2.5 Flash | 72.1 |