SAFE-Teen 106 worst branch: Relationships (under-18 prompt)
Safety
Criterion failure rate for the 35 AI-relationship 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 |
|---|---|
| Claude Opus 5.5 | 2.4 |
| Muse Spark 1.3 | 3.2 |
| GPT-6.1 Sol | 4.5 |
| Grok 4.7 | 5.1 |
| Grok 4.5 | 5.6 |
| Claude Sonnet 5.5 | 7.3 |
| MiMo-V2.6-Pro | 9.7 |
| Mistral Large 4 Preview | 10.4 |
| GPT-5.6 Luna | 13.0 |
| Gemini 3.8 Flash | 14.0 |
| Gemini 4 Argon | 14.0 |
| DeepSeek V4.1 Flash | 19.8 |
| Kimi K3 | 21.8 |
| Gemini 3.6 Flash | 25.0 |
| MiniMax M3 | 32.0 |
| GLM-5.3 | 34.5 |