OpenAI has launched a new speech architecture, GPT-Live, which supports real-time interaction and smooth background processing.

According to OneMillion AI monitoring, OpenAI has announced the rollout of the third-generation language model and architecture GPT-Live to global ChatGPT users. Free users can use GPT-Live-1 mini, while paid users will default to GPT-Live-1. This update separates complex reasoning from real-time conversation, allowing the model to listen while the user speaks without the need to take turns like a walkie-talkie.

GPT-Live features a full-duplex architecture, enabling bidirectional listening and speaking. During conversations, it uses interjections like "Hmm" and "Right" to indicate active listening, supporting rapid back-and-forth interactions or maintaining silence when the user needs time to think.

When faced with questions requiring search or deep reasoning, GPT-Live will offload the task to GPT-5.5 in the background and keep the conversation going while awaiting the results. It can also present visual cards for weather, stock market updates, and more to aid comprehension during the conversation.