OpenAI has released two new conversational voice models, GPT-Live-1 and GPT-Live-1 mini, which it says sound more natural and handle turn-taking more smoothly. The models are full-duplex, meaning they can listen and speak simultaneously, allowing users to interrupt naturally and enabling features such as live translation. OpenAI is replacing ChatGPT's existing Advanced Voice Mode with GPT-Live-1 mini by default, while paid subscribers gain access to the larger GPT-Live-1 model, marking a step towards the company's vision of voice as a primary computing interface.

Unlike the previous system, which chained together separate speech-to-text, language, and text-to-speech models, the new models route queries to OpenAI's latest text models such as GPT-5.5 for search, reasoning or agentic tasks while continuing the conversation. OpenAI says more than 150 million people already use features like Voice and Dictation, and its voice product lead, Atty Eleti, reported having 30- to 40-minute conversations with the feature. The company stressed the tool is not intended as an AI companion and includes safeguards for teenagers and sensitive topics, though the demo revealed shortcomings, including a heavy American accent and stilted, bookish phrasing during a Hindi live-translation test. Rivals including Apple, Amazon and startups such as Sesame and Monogram are pursuing similar, more expressive assistants.