SM Bench — Ambiguous Interpretation
Safety · 2026-02-01
SM Bench Ambiguous Interpretation tests whether a model asks for clarification when context is genuinely underspecified while avoiding needless refusal or an unwarranted harmful interpretation. Scores are difficulty-weighted judge credit across 100 fixed prompts.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 92.6 |
| Grok 4.5 | 92.0 |
| Sonnet 5 | 92.0 |
| GPT-5 Mini | 91.7 |
| GLM-5.1 | 90.8 |
| Gemma 4 31B IT | 90.8 |
| Opus 4.8 | 90.5 |
| GPT-5.1 | 90.2 |
| Kimi K2.6 | 89.9 |
| Opus 4.5 | 89.6 |
| Opus 4.7 | 89.3 |
| Kimi K2.5 | 89.3 |
| GLM-5 | 89.3 |
| Qwen3.6 Plus Preview | 89.3 |
| Sonnet 4.6 | 89.0 |
| GPT-5.2 | 89.0 |
| GPT-5.3 Instant | 88.7 |
| GPT-5.5 Instant | 88.1 |
| GPT-4.1 | 87.8 |
| GPT-5.4 | 87.8 |
| MiniMax M3 | 87.5 |
| GLM-5.2 | 87.2 |
| GPT-5.3-Codex | 86.9 |
| MiMo-V2.5-Pro | 86.6 |
| Haiku 4.5 | 86.6 |
| GPT-5.6 Terra Pro | 86.3 |
| GPT-5.4 Mini | 86.3 |
| GPT-5.4 Nano | 86.3 |
| Grok 4.3 | 86.0 |
| Gemini 3.5 Flash | 86.0 |
| GPT-5.6 Sol | 86.0 |
| MiniMax M2.7 | 86.0 |
| Gemini 3.1 Pro Preview | 85.7 |
| MiMo-V2-Pro | 85.1 |
| Gemini 3 Pro Preview | 84.5 |
| Opus 4.6 | 83.6 |
| GPT-5.6 Luna | 83.3 |
| Sonnet 4.5 | 82.7 |
| MiMo-V2-Omni | 82.7 |
| GPT-5.5 | 82.7 |