OpenAI has overhauled ChatGPT's voice mode with a new model, GPT-Live-1, which it says makes conversations feel more like "talking to another person". The key improvement is that the model is designed to interrupt users less and will pause to wait if a speaker breaks off mid-sentence, addressing a common frustration with the previous turn-based system. The move matters because voice is becoming a central way people interact with AI assistants, and more natural conversational flow could make the feature considerably more usable.

The new model is described as "full duplex", meaning it can listen and speak simultaneously, which enables features such as real-time translation as someone talks and the ability to tell the assistant to stay quiet until called upon. GPT-Live-1 can hand queries to OpenAI's stronger text models, such as GPT-5.5, for reasoning or web searches, and can display AI-generated visuals for topics like weather, stocks and sports. OpenAI says it has added safeguards steering the model away from harmful responses, offering vetted crisis helpline support and age-appropriate replies for teenagers — notable given the company faces lawsuits alleging ChatGPT harmed users' mental health. The model is rolling out on iOS, Android and the web, with a lighter "mini" version as the default for free users.