OpenAI Unveils Full-Duplex Voice Model GPT-Live
모델 출시 | Thu Jul 09 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 3 sources
OpenAI released GPT-Live-1 and its mini version, new voice models capable of real-time conversation.
Analysis
[OpenAI] released GPT-Live and GPT-Live-1 mini [1][2][3]
- Voice model based on full-duplex architecture
- Simultaneous listening and speaking
- Global rollout on iOS
- Android
- and web
- GPT-Live-1 for paid users
- mini for free users by default
[GPT-Live] equipped with natural conversation flow features [1][2][3]
- Backchanneling expressions like "mhmm" and "yeah"
- Ability to remain silent while the user is thinking
- Natural turn-taking without mid-speech interruptions
- Live translation support
[GPT-Live backend] handled reasoning via GPT-5.5 delegation [1][2][3]
- Delegates to GPT-5.5 when web search or deep reasoning is needed
- Maintains conversation flow during background tasks
- Will continue to be updated with the latest frontier models
- Provides AI-generated visuals for weather
- stocks
- and sports
[OpenAI safety measures] built in safeguards for high-risk situations [2]
- Built-in safeguards to prevent harmful responses
- Crisis helpline support during self-harm-related conversations
- Age-appropriate responses for minors
- Response to mental health-related lawsuit issues
[OpenAI] announced API release and voice-centric computing vision [1][3]
- API release planned for developers and enterprises
- Positions voice as the primary interface of computing
- Supports extended 30-40 minute conversations
- Over 150 million people use ChatGPT voice/dictation