SAFE-Teen 106 worst branch: Health Advice (under-18 prompt)
Safety
Criterion failure rate for the 36 health-advice scenarios in the 106-scenario cohort, with a teen-aware prompt, pooling the worst rating across each scenario's conversation branches for every applicable check. Judged denominators vary by model because blocked or unjudged responses are omitted.
Top models (lower is better)
| Model | Score |
|---|---|
| Muse Spark 1.3 | 4.5 |
| GPT-6.1 Sol | 7.7 |
| Grok 4.5 | 12.7 |
| GPT-5.6 Luna | 16.9 |
| Grok 4.7 | 19.2 |
| Claude Opus 5.5 | 27.7 |
| Claude Sonnet 5.5 | 30.8 |
| DeepSeek V4.1 Flash | 37.0 |
| Mistral Large 4 Preview | 38.7 |
| MiniMax M3 | 40.9 |
| MiMo-V2.6-Pro | 41.5 |
| Gemini 4 Argon | 43.6 |
| Gemini 3.6 Flash | 46.9 |
| Gemini 3.8 Flash | 48.1 |
| Kimi K3 | 49.1 |
| GLM-5.3 | 50.9 |