OpenAI Replaces Advanced Voice Mode With Full-Duplex GPT-Live-1
OpenAI launched GPT-Live-1 and GPT-Live-1 mini, voice models that speak and listen at once and route complex queries to its reasoning models. The mini version is now ChatGPT's default voice.
OpenAI shipped GPT-Live-1 and GPT-Live-1 mini on July 8, retiring Advanced Voice Mode in favor of full-duplex voice models that speak and listen at the same time.
What full-duplex actually fixes
Advanced Voice Mode, like most voice AI, worked in turns: you talk, it processes, it responds, you wait your turn to interrupt. Full-duplex means GPT-Live-1 processes incoming audio while it’s still generating output, so a user can cut in mid-sentence and the model actually reacts instead of finishing its canned response first. OpenAI is also positioning it for live-translation-style use cases, where continuous bidirectional audio matters more than turn-based exchange.
This is a separate release from the GPT-5.6 Sol/Terra/Luna text-model rollout OpenAI announced earlier the same week — GPT-Live-1 is a voice layer that sits on top of the reasoning stack, not a new base model in its own right.
How the routing works
GPT-Live-1 mini becomes ChatGPT’s new default voice, replacing Advanced Voice Mode for all users. Paid-tier subscribers get access to the larger GPT-Live-1. Under the hood, the voice layer routes queries needing search or multi-step reasoning to OpenAI’s latest text models — the voice model handles the conversational, low-latency audio loop, and hands off to the heavier reasoning stack when a query actually needs it.
That’s a meaningful architectural admission: real-time voice and deep reasoning are still separate problems that get solved by separate models, stitched together rather than merged into one.
Scale and rollout
OpenAI said more than 150 million people already use ChatGPT’s voice and dictation features — a number that puts this squarely in “change the default experience for a mass audience” territory rather than a niche feature update. Rollout is global, hitting iOS, Android, CarPlay, and ChatGPT.com simultaneously.
Why it matters
The complaint list OpenAI is answering — clumsy interruption handling, a voice mode that felt noticeably dumber than the text chat — has been a recurring criticism since Advanced Voice Mode launched. Full-duplex audio plus routing to frontier reasoning models closes both gaps at once, assuming the interruption handling holds up outside a demo. With Gemini and Grok both pushing multimodal, low-latency voice as a wedge into daily assistant use, this is OpenAI defending ground it can’t afford to cede on a feature 150 million people already touch weekly.