SM Bench — NSFW (Explicit)
Safety · 2026-02-01
SM Bench NSFW (Explicit) tests default model behaviour on explicit adult-content requests according to the benchmark's permissive usefulness rubric. Scores are difficulty-weighted judge credit across 100 fixed prompts.
Top models (higher is better)
| Model | Score |
|---|---|
| Grok 4.5 | 100.0 |
| Grok 4.6 | 100.0 |
| Grok 4.3 | 100.0 |
| Grok 4.7 | 100.0 |
| Grok 4.1 Fast | 100.0 |
| Claude Fable 5.1 | 89.5 |
| Kimi K2.5 | 80.3 |
| DeepSeek-V4-Pro | 79.1 |
| LongCat-2.0 | 67.2 |
| Gemini 3 Flash Preview | 66.7 |
| DeepSeek-V3.2 | 64.2 |
| GLM-5.1 | 62.2 |
| Mistral Small 4 | 60.5 |
| Kimi K2.6 | 55.2 |
| Sonnet 4.5 | 54.2 |
| Opus 4.7 | 54.0 |
| Muse Glimmer 30B | 51.2 |
| Claude Sonnet 5.5 | 48.0 |
| Gemini 3.1 Pro Preview | 45.8 |
| Opus 4.5 | 44.3 |
| Opus 4.6 | 39.8 |
| GLM-5.2 | 38.8 |
| Claude Opus 5 | 37.8 |
| GLM 5 Turbo | 35.8 |
| GLM-5 | 34.3 |
| Grok 4.20 | 30.4 |
| MiMo-V2.6-Pro | 27.4 |
| DeepSeek-V4-Flash | 24.4 |
| Gemini 3.1 Flash-Lite Preview | 24.1 |
| GPT-4o Mini | 19.1 |
| Trinity Large Preview | 18.9 |
| Claude Opus 5.5 | 17.7 |
| Gemini 3 Pro Preview | 17.2 |
| MiMo-V2-Omni | 17.2 |
| Qwen3.7-Max (source unspecified) | 16.9 |
| Trinity Large Thinking | 15.4 |
| Muse Spark 1.1 | 14.9 |
| GLM-4.7 | 14.4 |
| Claude Haiku 5.5 | 13.7 |
| MiMo-V2.6-Flash | 12.4 |