OpenAI is trying to make ChatGPT’s voice mode feel less like a walkie-talkie and more like a live conversation.
The company introduced GPT-Live-1 on Wednesday, July 8, a new voice model designed to listen while it speaks, interrupt users less and wait when someone pauses mid-thought. The feature is rolling out across iOS, Android and the web.
OpenAI research lead Kundan Kumar called GPT-Live-1 the company’s “smartest voice model” during a press briefing. The model can also hand more complex requests to stronger text models, including GPT-5.5, when it needs to reason through an answer or search the web.
The biggest change is the move away from an older turn-based system, where users and the chatbot had to take turns more rigidly. OpenAI product lead Atty Eleti said GPT-Live-1 is a “full duplex model,” meaning it can process incoming speech and produce spoken responses at the same time.
That lets ChatGPT translate speech in real time while a person is still talking. Users can also tell ChatGPT Voice to stay quiet until called on, with the assistant acknowledging that it is listening through short verbal cues such as “mhmm” or “got it.”
OpenAI said the voice upgrade can add AI-generated visuals to conversations about weather, stocks and sports, including forecasts or scores. The company also said it has added safeguards intended to move the model away from harmful replies or end conversations in higher-risk situations.
The safety claims arrive as OpenAI faces lawsuits alleging ChatGPT harmed some users’ mental health. OpenAI said the new model is trained to offer expert-vetted crisis helpline support in self-harm conversations and provide age-appropriate responses for teens.
GPT-Live-1 will power ChatGPT Voice for Go, Plus and Pro subscribers, while free users will get a smaller GPT-Live-1 mini model by default.
