Benchmarks
347 sources · 1,258 models × 3,038 benchmarks · 158,912 scoresVals clinician-authored safety evaluation spanning 101 scenarios for adolescents aged 13–17. Simulated branching conversations use GPT-5.6 Terra and are judged by Claude Sonnet 5.5. Risk-group failure rates pool judged criteria, excluding blocked or otherwise unjudged responses; they are not independent scenario pass rates. No-prompt and under-18 system-prompt conditions are separate.
No signal or difficulty estimate. This benchmark category is excluded from the capability model and its validation studies.
This is a benchmark-family page. Scores are recorded on the component benchmarks below rather than directly on the parent, because the components use different score scales and directions.
| Metric | Models | Leader | Direction | Date |
|---|---|---|---|---|
| SAFE-Teen Health Advice (no prompt) | 10 | GPT-6.1 Sol4.2 | lower is better | 2026-10-08 |
| SAFE-Teen Health Advice (under-18 prompt) | 10 | Muse Spark 1.30.8 | lower is better | 2026-10-08 |
| SAFE-Teen Relationships (no prompt) | 10 | GPT-6.1 Sol10.0 | lower is better | 2026-10-08 |
| SAFE-Teen Relationships (under-18 prompt) | 10 | Claude Opus 5.50.5 | lower is better | 2026-10-08 |
| SAFE-Teen Urgent Safety (no prompt) | 10 | GPT-6.1 Sol3.5 | lower is better | 2026-10-08 |
| SAFE-Teen Urgent Safety (under-18 prompt) | 10 | GPT-6.1 Sol2.1 | lower is better | 2026-10-08 |