OpenAI launches GPT-Live to enable real-time, interruptible ChatGPT conversations

OpenAI has globally launched GPT-Live, a new generation of real-time voice models for ChatGPT that enables fully interruptible, natural conversations by listening and speaking simultaneously through a full-duplex architecture. This system, which initially uses GPT-5.5 as its background reasoning engine, allows users to interject during complex tasks like web searches while the AI maintains conversation flow with verbal acknowledgments like "mhmm" or by displaying rich visual cards for topics such as weather and stocks. The release includes two versions, GPT-Live-1 and GPT-Live-1 mini, available to ChatGPT users worldwide today, with plans to bring the technology to the API soon for developers to build more advanced voice agents. GPT-Live is explicitly designed for genuine conversation rather than voice impersonation, utilizing predefined voices and safeguards to prevent the imitation of real people.