SM Bench — EQ Boundaries
Safety · 2026-02-01
SM Bench EQ Boundaries tests emotionally intelligent responses that remain supportive and proportionate without unnecessary clinical distancing, canned disclaimers, or inappropriate dependency reinforcement. Scores are difficulty-weighted judge credit across 100 fixed prompts.
Top models (higher is better)
| Model | Score |
|---|---|
| Grok 4.20 | 90.5 |
| Mistral Small 4 | 85.7 |
| MiMo-V2-Pro | 79.8 |
| GPT-6.1 Sol | 78.1 |
| GPT-6 Astra | 77.0 |
| GPT-6 Astra Pro | 76.4 |
| Gemini 3.5 Flash-Lite | 76.1 |
| Grok 4.1 Fast | 74.7 |
| GPT-6 Sol | 74.7 |
| Grok 4.6 | 74.2 |
| MiMo-V2-Omni | 74.2 |
| GLM-5.1 | 73.9 |
| GPT-4.1 | 73.3 |
| Gemini 3.6 Flash | 71.3 |
| GLM-4.7 | 71.3 |
| GPT-4o Mini | 71.3 |
| Grok 4.3 | 70.8 |
| LongCat-2.0 | 70.8 |
| Trinity Large Preview | 70.8 |
| Kimi K3 | 70.5 |
| MiMo-V2.5 | 70.5 |
| DeepSeek-V4-Flash | 69.9 |
| GLM-5.3 | 69.7 |
| Grok 4.5 | 69.1 |
| Grok 4.7 | 69.1 |
| GPT-5.3-Codex | 69.1 |
| GLM 5 Turbo | 69.1 |
| Claude Opus 5 | 68.8 |
| MiniMax M2.1 | 68.8 |
| Gemini 3.5 Flash | 68.5 |
| Gemini 3.1 Pro Preview | 68.3 |
| Muse Spark 1.1 | 68.3 |
| GPT-5.4 Mini | 68.3 |
| MiMo-V2.5-Pro | 68.0 |
| Claude Fable 5.1 | 67.7 |
| Gemma 4 31B IT | 67.7 |
| DeepSeek-V3.2 | 67.4 |
| Muse Spark 1.2 | 67.4 |
| Gemini 3.1 Flash-Lite Preview | 66.8 |
| GPT-5.6 Terra Pro | 66.6 |