OpenAI Retires Advanced Voice Mode for GPT-Live, ChatGPT Now Listens and Talks at Once
TL;DR
OpenAI launched full-duplex voice model GPT-Live on July 8, replacing Advanced Voice Mode for 150M ChatGPT voice users, with GPT-5.5 running in the background.
OpenAI released GPT-Live on July 8, retiring the two-year-old Advanced Voice Mode inside ChatGPT and replacing it with a full-duplex voice architecture. The old pipeline chained speech-to-text, an LLM and text-to-speech, forcing strict turn-taking. GPT-Live folds the three stages into a single model that can speak and listen at the same time, drop in a "mhmm" or "got it" mid-sentence, or stay silent while the user thinks.
Two variants shipped: GPT-Live-1 is the default for Go, Plus and Pro subscribers, and GPT-Live-1 mini is the default voice model for free users worldwide. ChatGPT Voice product lead Atty Eleti told TechCrunch, "Over time, we think this will also unlock the ability to use voice as a kind of primary interface to computing." OpenAI said ChatGPT voice features now have 150 million monthly users, and its demos ran continuous conversations of 30 to 40 minutes.
The twist is that this voice-first product outsources the hard work. When a question needs web search, deeper reasoning or tool use, GPT-Live hands the request to GPT-5.5 in the background and pipes the result back into the conversation stream, while the front-end model handles only acoustics and timing. A live Hindi translation demo carried a heavy American accent and unnatural phrasing; OpenAI said non-English languages are still being tuned. The API is not out yet, and developers have to join a waitlist.
Nine remastered voices, rich cards for weather, stocks and sports, and CarPlay integration all shipped alongside. Video and screen sharing are "coming soon." Google DeepMind and xAI pushed their own realtime voice agents the same week. This launch takes the Her-style GPT-4o demo Sam Altman ran in May last year and moves it into the main product line.
via OpenAI / TechCrunch / MacRumors
Two variants shipped: GPT-Live-1 is the default for Go, Plus and Pro subscribers, and GPT-Live-1 mini is the default voice model for free users worldwide. ChatGPT Voice product lead Atty Eleti told TechCrunch, "Over time, we think this will also unlock the ability to use voice as a kind of primary interface to computing." OpenAI said ChatGPT voice features now have 150 million monthly users, and its demos ran continuous conversations of 30 to 40 minutes.
The twist is that this voice-first product outsources the hard work. When a question needs web search, deeper reasoning or tool use, GPT-Live hands the request to GPT-5.5 in the background and pipes the result back into the conversation stream, while the front-end model handles only acoustics and timing. A live Hindi translation demo carried a heavy American accent and unnatural phrasing; OpenAI said non-English languages are still being tuned. The API is not out yet, and developers have to join a waitlist.
Nine remastered voices, rich cards for weather, stocks and sports, and CarPlay integration all shipped alongside. Video and screen sharing are "coming soon." Google DeepMind and xAI pushed their own realtime voice agents the same week. This launch takes the Her-style GPT-4o demo Sam Altman ran in May last year and moves it into the main product line.
via OpenAI / TechCrunch / MacRumors
