OpenAI's New Voice Models Handle Interruptions Like a Human

OpenAI just dropped two new conversational models, GPT-Live-1 and GPT-Live-1 mini, designed to sound way more natural. The big deal here is full-duplex audio — they can talk and listen at the same time — so you can interrupt them without awkward pauses. This replaces the old Advanced Voice Mode in ChatGPT, which relied on a clunky pipeline of separate models for speech-to-text, LLM response, and text-to-speech. The mini version is now the default for ChatGPT free users, while paid tiers get the full model. These new models also tap into GPT-5.5 for reasoning and search, meaning they can stay quiet for a while, soak up context, and even show visuals when needed. OpenAI claims people are already having 30-40 minute voice conversations, and they see voice as the future primary interface for complex work. Rivals like Apple, Amazon, and startups like Sesame and Monogram are racing to make assistants more expressive too. Why it matters: This shift to natural voice interaction could fundamentally change how we use AI, moving from typing prompts to hands-free, ongoing conversations that feel less like talking to a robot and more like chatting with a capable colleague.