OpenAI has unveiled GPT-Live, a groundbreaking voice AI that enhances conversational fluidity by speaking and listening simultaneously. This advanced system is built on a full-duplex framework, allowing it to respond in real-time, even while users are still speaking. By incorporating natural conversational cues like “mhmm” and “yeah,” GPT-Live aims to facilitate more seamless and faster interactions, minimizing awkward pauses.
The innovation doesn’t stop there; for more intricate requests requiring web searches or complex reasoning, GPT-Live can delegate these tasks to a more robust AI model operating in the background, all while maintaining the conversation. Currently, these complex tasks are handled by GPT-5.5, with plans to support more advanced models in upcoming updates.
This announcement coincides with OpenAI’s recent confirmation that the GPT-5.6 model series will soon be available to the public, pending the completion of additional cybersecurity evaluations. The GPT-5.6 lineup features the Sol model as its centerpiece, accompanied by variants named Terra and Luna.
As part of the rollout, OpenAI has released two iterations of the new voice models—GPT-Live-1 and GPT-Live-1 mini—making them accessible to ChatGPT users globally. Additionally, the company intends to offer GPT-Live through its API, enabling developers and businesses to integrate real-time voice AI capabilities into their applications.