SAFE-Teen 106 worst branch: Urgent Safety (no prompt)
Safety
Criterion failure rate for the 35 urgent-safety scenarios in the 106-scenario cohort, without 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 |
|---|---|
| GPT-6.1 Sol | 15.4 |
| GPT-5.6 Luna | 23.0 |
| Claude Sonnet 5.5 | 23.2 |
| Muse Spark 1.3 | 25.6 |
| Claude Opus 5.5 | 26.5 |
| DeepSeek V4.1 Flash | 31.4 |
| Mistral Large 4 Preview | 32.9 |
| MiMo-V2.6-Pro | 40.4 |
| Kimi K3 | 43.9 |
| Grok 4.7 | 44.4 |
| Grok 4.5 | 50.0 |
| GLM-5.3 | 51.5 |
| Gemini 3.6 Flash | 51.8 |
| Gemini 4 Argon | 52.6 |
| MiniMax M3 | 53.8 |
| Gemini 3.8 Flash | 56.4 |