OpenAI released two new conversational models on Tuesday, TechCrunch reports — GPT-Live-1 and GPT-Live-1 mini — that can speak and listen simultaneously, letting users interrupt naturally and enabling live translation. ChatGPT's Advanced Voice Mode switches to GPT-Live-1 mini by default; paid users unlock the full model.
The new system fixes longstanding complaints: the old voice mode would cut people off mid-sentence and lacked the intelligence for complex questions. Now queries route to GPT-5.5 for search and reasoning while the conversation continues uninterrupted. "We think voice can be the future interface to all kinds of work," said product lead Atty Eleti, who says he regularly has 40-minute walking conversations with the feature.
Over 150 million people already talk to ChatGPT through Voice and Dictation. OpenAI stressed the new models aren't designed as AI companions — they include age-appropriate safeguards for teens and resources for sensitive topics like self-harm.
Analysis
Full-duplex voice is the missing piece for ambient computing. If OpenAI can make conversation feel natural enough that people talk to it for 40 minutes unprompted, the earbuds rumoured for later this year start to make sense. But the Hindi demo's heavy American accent shows the localisation problem isn't solved yet — and that's table stakes for a global voice platform.
