Skip to main content
BenchLM
Data
Official provider results

GPT-Live-1.1 system card safety evaluations

OpenAI’s GPT-Live-1.1 system card reports voice-native safety scores for GPT-Live-1.1 API against GPT-Live-1 API, on real shared voice conversations and on synthetic spoken edge cases. OpenAI published no capability benchmarks, such as task success or latency, for GPT-Live-1.1.

Source snapshots refreshed
Five measurement lanesVOICE / S2S

Reasoning, task completion, conversation dynamics, experience, and latency stay separate.

Back to directory

Benchmark profile

Scope
Six disallowed-content categories on production voice prompts and eight on synthetic spoken prompts; both API models run with delegation disabled
Primary metric
Share of responses graded as not unsafe, from 0 to 1 (higher is better)
Owner
OpenAI
Available evidence
Exact figures in tables 1 and 2 of the official system card PDF

What it measures

Conversation dynamics
Can it handle turns, interruptions, ambiguity, and state?

Available results

Production prompts: sexual
0.66
Same as GPT-Live-1 API

GPT-Live-1 API scores 0.66 on the same production prompts set.

Production prompts: illicit behavior
0.78
Same as GPT-Live-1 API

GPT-Live-1 API scores 0.78 on the same production prompts set.

Production prompts: mental health
0.62
Above GPT-Live-1 API

GPT-Live-1 API scores 0.61 on the same production prompts set.

Production prompts: personal data
0.88
Same as GPT-Live-1 API

GPT-Live-1 API scores 0.88 on the same production prompts set.

Production prompts: emotional reliance
0.72
Above GPT-Live-1 API

GPT-Live-1 API scores 0.65 on the same production prompts set.

Production prompts: self-harm
0.78
Above GPT-Live-1 API

GPT-Live-1 API scores 0.77 on the same production prompts set.

Synthetic prompts: sexual
0.89
Above GPT-Live-1 API

GPT-Live-1 API scores 0.87 on the same synthetic prompts set.

Synthetic prompts: illicit behavior
0.92
Above GPT-Live-1 API

GPT-Live-1 API scores 0.89 on the same synthetic prompts set.

Synthetic prompts: mental health
0.70
Above GPT-Live-1 API

GPT-Live-1 API scores 0.67 on the same synthetic prompts set.

Synthetic prompts: personal data
0.93
Above GPT-Live-1 API

GPT-Live-1 API scores 0.92 on the same synthetic prompts set.

Synthetic prompts: emotional reliance
0.85
Above GPT-Live-1 API

GPT-Live-1 API scores 0.82 on the same synthetic prompts set.

Synthetic prompts: self-harm
0.85
Above GPT-Live-1 API

GPT-Live-1 API scores 0.82 on the same synthetic prompts set.

Synthetic prompts: hate
0.97
Above GPT-Live-1 API

GPT-Live-1 API scores 0.95 on the same synthetic prompts set.

Synthetic prompts: gore
0.85
Above GPT-Live-1 API

GPT-Live-1 API scores 0.74 on the same synthetic prompts set.

Interpretation limit

These are provider-reported safety results, not a capability or voice-quality score. OpenAI built the sets around cases where earlier voice models gave poor answers, so the scores are not prevalence-weighted and do not describe real traffic. The API models were run with delegation off, so they show the front-end voice model alone, not the backend model a deployment would use. The ChatGPT configurations of GPT-Live-1.1 and GPT-Live-1.1 Quip are scored separately in the card and are not shown here.