OpenAI announced the launch of GPT-Live, a new generation of voice models designed to make talking with AI feel like a real conversation. Built on a full-duplex architecture, GPT-Live can listen and speak at the same time, allowing for quick back-and-forth, natural interruptions, and continuous interaction without relying on discrete turns dictated by silence. The system decouples continuous conversation management from deeper reasoning tasks, enabling GPT-Live to delegate complex work requiring web search or reasoning to background frontier models like GPT-5.5 while maintaining the flow of conversation. OpenAI is rolling out two versions, GPT-Live-1 and GPT-Live-1 mini, globally to ChatGPT users, with upcoming availability planned for developers and enterprises via the API. Additionally, OpenAI announced that supported audio generated with GPT-Live through ChatGPT Voice and the API now includes SynthID watermarking, accompanied by a public verification tool and API access for provenance checks. In human evaluations, GPT-Live models outperformed Advanced Voice Mode across overall preference, conversational flow, and specialized benchmarks testing expert-level reasoning, web search, and telecom support tasks.
Introducing GPT-Live
OpenAI has announced GPT-Live, a new generation of voice models featuring a full-duplex architecture designed for continuous interaction and natural human-AI conversation. Rolled out globally in two versions, GPT-Live-1 and GPT-Live-1 mini, the models power an updated ChatGPT Voice experience and will soon be available via the API.
Why it matters
The transition to full-duplex voice models overcomes the rigid turn-taking limitations of earlier cascaded and turn-based voice AI systems, moving the industry closer to fluid, real-time human-computer interaction.
Topic dossiers OpenAI