Self-harm
We haven't written a description for this benchmark yet; see the card notes below for how each maker ran it.
- ModelScoreYour cost / mo
- 1GPT-Live 1 · OpenAI0.97 score
- 2GPT-Live 1 · OpenAI0.96 score
“Best value”: no other model here reports a higher score at a lower cost for your workload. Scores are self-reported by each maker with different settings, so treat close calls as ties.
How each maker reported it
- GPT-Live 1 system card §3.1 (Tables 1–2, corrected 4 Aug 2026)
Safety scores on OpenAI's voice-native disallowed-content evaluations (higher is safer). Prompts were chosen where the earlier Advanced Voice Mode (AVM) models did badly, so these are not production rates. Evaluated with delegation, as deployed in ChatGPT. OpenAI published no capability benchmarks for GPT-Live-1.
- GPT-Live 1 system card §3.1 (Tables 1–2, corrected 4 Aug 2026)
Safety scores on OpenAI's voice-native disallowed-content evaluations (higher is safer). Prompts were chosen where the earlier Advanced Voice Mode (AVM) models did badly, so these are not production rates. Evaluated with delegation, as deployed in ChatGPT. OpenAI published no capability benchmarks for GPT-Live-1.
Only each model's own maker's number is shown; scores a maker reports for rival models are left out.