Phare Tools Reliability
Safety · 2025-03-27
Phare task testing reliability when using or reasoning about tool/plugin outputs.
Top models (higher is better)
| Model | Score |
|---|---|
| Sonnet 4.6 | 95.5 |
| Opus 4.6 | 94.3 |
| Gemini 3.1 Pro Preview | 94.2 |
| DeepSeek-V4-Pro | 93.9 |
| Sonnet 5 | 93.6 |
| Gemini 3.5 Flash | 93.3 |
| Opus 4.5 | 93.1 |
| DeepSeek-V4-Flash | 92.5 |
| Sonnet 4.5 | 92.2 |
| GLM-5.2 | 91.9 |
| Qwen3.7 Max Preview | 91.6 |
| Opus 4.1 | 91.2 |
| Kimi K2.5 | 90.3 |
| Claude 3.5 Sonnet (Oct 2024) | 89.9 |
| Grok 4 | 89.8 |
| Qwen3.7-Plus | 89.7 |
| GPT-5.5 | 89.2 |
| Gemini 3 Pro Preview | 89.1 |
| Gemma 4 31B IT | 87.7 |
| Gemini 3.1 Flash-Lite | 87.6 |
| Grok 4.3 | 85.9 |
| Haiku 4.5 | 84.8 |
| GPT-4.1 | 84.5 |
| Mistral Medium 3.5 | 84.4 |
| Kimi K2.6 | 83.9 |
| Grok 3 | 83.2 |
| GPT-5.1 | 82.8 |
| Mistral Large 3 675B Instruct 2512 | 82.6 |
| Haiku 3.5 | 82.6 |
| Mistral Large 2 (Instruct 2407) | 82.5 |
| Mistral Medium 3.1 | 82.4 |
| GPT-4.1 Mini | 82.1 |
| GPT-4o | 81.7 |
| GPT-5 Mini | 81.6 |
| Claude 3.7 Sonnet | 81.2 |
| Grok 2 | 81.0 |
| GPT-5.2 | 80.8 |
| DeepSeek-V3 | 80.2 |
| GPT-4o Mini | 79.8 |
| gpt-oss-120b | 79.3 |