EQ-Bench 3: Safety
Chat & Writing · 2025-04-28
Within EQ-Bench 3's emotional-intelligence scenarios, this subscore rates the safety dimension of model responses on a 0–10 scale. Higher values are better.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.2 | 7.7 |
| Grok 4.20 | 7.7 |
| Horizon Alpha | 7.7 |
| GPT-5.5 | 7.6 |
| GPT-5.4 | 7.5 |
| Kimi K2.5 | 7.3 |
| o3 | 7.2 |
| Kimi K2.6 | 7.0 |
| Kimi K2 Instruct | 7.0 |
| Claude Fable 5 | 6.9 |
| Opus 4.6 | 6.9 |
| Opus 4.8 | 6.7 |
| Sonnet 4 | 6.7 |
| Sonnet 4.6 | 6.7 |
| Hivemind-32B-Preview | 6.7 |
| Qwen3.5 397B A17B | 6.7 |
| Claude 3.7 Sonnet | 6.6 |
| Sonnet 4.5 | 6.6 |
| Gemini 3.1 Pro Preview | 6.6 |
| Gemini 3 Pro Preview | 6.6 |
| Gemma 4 26B A4B IT | 6.6 |
| GPT-4.1 | 6.6 |
| GPT-5.1 | 6.6 |
| gpt-oss-120b | 6.6 |
| Gemma 4 31B IT | 6.5 |
| GLM-4.7 | 6.5 |
| Opus 4.5 | 6.4 |
| Opus 4.7 | 6.4 |
| GLM-5.1 | 6.4 |
| Opus 4 | 6.3 |
| GLM-5.2 | 6.3 |
| Gemma 3 27B IT | 6.2 |
| DeepSeek-V4-Pro | 6.1 |
| GLM-4.5 | 6.1 |
| O4 Mini | 6.1 |
| Claude 3.5 Sonnet (Oct 2024) | 6.0 |
| DeepSeek-R1 | 6.0 |
| GPT-5 Chat (2025-08-07) | 6.0 |
| Gemini 2.5 Pro Preview 05-06 | 5.9 |
| GLM-5 | 5.9 |