SM Bench — Overfit
Safety · 2026-02-01
SM Bench Overfit tests whether surface-level risk cues and policy-shaped phrasing override benign intent, common-sense reasoning, or useful task completion. Scores use difficulty-weighted pass, partial, and fail judgments across 100 fixed prompts; this category receives 2× weight in the overall SM Bench score.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 3.5 Flash | 98.4 |
| Gemini 3.1 Flash-Lite Preview | 97.8 |
| MiMo-V2.6-Pro | 96.7 |
| Opus 4.6 | 95.6 |
| Gemini 3.6 Flash | 95.1 |
| Gemini 3.8 Flash | 95.1 |
| GPT-6 Astra | 95.1 |
| GPT-6 Astra Pro | 94.5 |
| Gemini 3 Flash Preview | 94.0 |
| Gemini 3.7 Flash | 94.0 |
| Kimi K3 | 94.0 |
| GPT-6.1 Sol | 93.2 |
| Grok 4.5 | 92.3 |
| GLM-5.3 | 92.3 |
| Opus 4.7 | 91.8 |
| Gemma 4 31B IT | 91.8 |
| Sonnet 4.6 | 91.5 |
| Claude Opus 5 | 91.3 |
| MiMo-V2-Pro | 90.7 |
| MiMo-V2.6-Flash | 90.7 |
| Claude Sonnet 5.5 | 89.6 |
| Claude Opus 5.5 | 89.1 |
| Opus 4.5 | 88.5 |
| Muse Spark 1.1 | 88.0 |
| Opus 4.8 | 87.4 |
| MiMo-V2.5-Pro | 85.8 |
| GPT-6 Sol | 85.5 |
| Claude Fable 5.1 | 84.7 |
| Gemini 3.1 Pro Preview | 84.2 |
| Trinity Large Preview | 83.9 |
| Sonnet 4.5 | 83.1 |
| Gemini 3 Pro Preview | 83.1 |
| Sonnet 5 | 83.1 |
| GPT-4o (source unspecified) | 83.1 |
| Haiku 4.5 | 82.0 |
| MiMo-V2-Omni | 81.4 |
| Grok 4.6 | 80.9 |
| Grok 4.3 | 80.9 |
| GLM-5.2 | 80.3 |
| Muse Spark 1.2 | 80.3 |