OpenAI launched two new voice models, GPT-Live-1 and GPT-Live-1 mini, on August 6, 2026, enabling full duplex conversations that let the AI speak and listen simultaneously on iOS, Android, and web platforms [1, 2, 3, 4]. GPT-Live replaces the previous turn-based Advanced Voice Mode by allowing users to interrupt the AI at any time and supports real-time translation while speaking [1, 5, 6]. "GPT-Live-1 is the company’s smartest voice model yet," said OpenAI researcher lead Kundan Kumar [1].
GPT-Live's ability to produce short verbal acknowledgments like "mhmm" or "yeah" during conversations makes the interactions feel more natural and human-like [1, 7]. The models route complex queries that need reasoning or web searches to OpenAI’s more advanced text models, such as GPT-5.5, running in the background for smooth processing [1, 5]. OpenAI president Greg Brockman called it "a much more natural way of interacting with your computer" [6].
The voice models come with built-in safeguards to avoid harmful content, provide crisis helpline information, and offer age-appropriate responses for teens [1, 3, 4]. They can also generate visual cards showing weather, sports scores, and stock information during conversations [1, 5]. Benchmark tests show GPT-Live has improved scientific reasoning and web search capabilities compared to earlier voice models [8].
Before launching GPT-Live, OpenAI reported more than 150 million users weekly accessed ChatGPT's voice and dictation features [5]. Conversations with GPT-Live have been tested up to 30-40 minutes, with blind testing flows typically around 5-10 minutes to measure comfort and naturalness [5, 3, 4].
OpenAI plans to integrate GPT-Live voice functions deeply into its APIs in the coming weeks for developers and enterprises, although voice support for video or screen-sharing is not yet available [3, 4]. Product lead Atty Eleti said, "Over time, we think this will also unlock the ability to use voice as a kind of primary interface to computing, and to manage increasingly complex long-running agentic work."
