SM Bench
Safety · 2026-02-01
Safetymaxxed Bench (SM Bench) evaluates safety overfitting and useful alignment on a fixed v1 battery of 800 prompts across eight categories. Gemini 3 Flash Preview judges each response as pass, partial, or fail; item credit is weighted 1×, 2×, or 3× for easy, medium, or hard cases, and the overall score is the category-weighted mean with Overfit weighted 2×. Higher is better.
Top models (higher is better)
| Model | Score |
|---|---|
| Grok 4.5 | 92.1 |
| Claude Fable 5.1 | 89.0 |
| Grok 4.6 | 89.0 |
| Grok 4.3 | 88.9 |
| Grok 4.7 | 86.3 |
| Claude Opus 5 | 85.8 |
| Claude Sonnet 5.5 | 85.4 |
| Gemini 3 Flash Preview | 85.3 |
| Grok 4.1 Fast | 84.8 |
| GLM-5.1 | 84.1 |
| Opus 4.7 | 84.0 |
| Gemini 3.1 Flash-Lite Preview | 83.8 |
| Gemini 3.1 Pro Preview | 83.4 |
| Kimi K2.5 | 82.3 |
| Gemma 4 31B IT | 82.2 |
| MiMo-V2.6-Pro | 82.0 |
| Gemini 3.5 Flash | 82.0 |
| Claude Opus 5.5 | 81.8 |
| GLM-5.3 | 81.6 |
| MiMo-V2-Pro | 81.5 |
| DeepSeek-V4-Pro | 81.5 |
| Opus 4.5 | 81.4 |
| GLM-5.2 | 81.3 |
| Gemini 3.6 Flash | 80.8 |
| Opus 4.6 | 80.8 |
| Gemini 3.7 Flash | 80.6 |
| Gemini 3.8 Flash | 80.4 |
| Sonnet 4.5 | 79.8 |
| Gemini 3 Pro Preview | 79.3 |
| MiMo-V2-Omni | 79.2 |
| LongCat-2.0 | 78.9 |
| Kimi K2.6 | 78.5 |
| Kimi K3 | 77.8 |
| Gemini 3.5 Flash-Lite | 77.4 |
| GPT-4.1 | 77.3 |
| GLM-5 | 77.0 |
| Qwen3.7-Max (source unspecified) | 76.8 |
| MiMo-V2.6-Flash | 76.6 |
| Muse Spark 1.1 | 75.9 |
| Qwen3.6 Plus Preview | 74.6 |