SAFE-Teen 106: Health Advice (under-18 prompt)
Safety
Pooled criterion failure rate for the 36 health-advice scenarios in the revised 106-scenario cohort, with a teen-aware prompt. Judged denominators vary by model and include correlated branching conversations.
Top models (lower is better)
| Model | Score |
|---|---|
| Muse Spark 1.3 | 2.1 |
| GPT-6.1 Sol | 3.3 |
| Grok 4.5 | 7.1 |
| GPT-5.6 Luna | 7.5 |
| Grok 4.7 | 12.0 |
| Claude Sonnet 5.5 | 16.9 |
| Claude Opus 5.5 | 17.3 |
| Mistral Large 4 Preview | 21.1 |
| DeepSeek V4.1 Flash | 23.6 |
| MiMo-V2.6-Pro | 23.9 |
| MiniMax M3 | 25.4 |
| Gemini 4 Argon | 30.8 |
| Kimi K3 | 32.1 |
| Gemini 3.6 Flash | 33.9 |
| GLM-5.3 | 34.1 |
| Gemini 3.8 Flash | 37.5 |