OpenAI has unveiled GPT-Live, an innovative voice AI designed to enhance the natural flow of conversations by enabling simultaneous speaking and listening. This advancement is built on a full-duplex architecture, allowing the AI to acknowledge user input with conversational cues like “mhmm” or “yeah,” which helps maintain a more fluid dialogue without the typical long pauses.
In instances where a more intricate response is necessary, such as those requiring web searches or complex reasoning, GPT-Live can effortlessly delegate the task to a more robust AI model in the background, all while maintaining the conversation uninterrupted. These sophisticated tasks are initially managed by GPT-5.5, with plans to incorporate support for cutting-edge models in upcoming updates.
The introduction of GPT-Live comes on the heels of OpenAI’s announcement regarding the forthcoming public release of its GPT-5.6 model series. This series includes the prominent Sol model, alongside Terra and Luna variants, and will be made available following additional cybersecurity assessments and approvals.
OpenAI has started deploying two versions of these novel voice models—GPT-Live-1 and GPT-Live-1 mini—to ChatGPT users around the globe. Furthermore, the company intends to provide access to GPT-Live through its API, which will allow developers and businesses to integrate real-time voice AI functionalities into their applications.