On July 8, OpenAI unveiled GPT-Live — a new generation of voice models that replaces Advanced Voice Mode as the engine for ChatGPT Voice. The technical core is what OpenAI calls a full-duplex architecture.
What ‘full-duplex’ means
Until now, voice AI ran as a cascade: transcription, then a language model, then synthesis — with a turn-taking model waiting for a clean silence before responding. GPT-Live instead processes your audio continuously, even while it’s speaking. It can drop in short backchannels (‘mhmm’, ‘okay’) while you’re still talking, and decides several times per second whether to speak, listen, pause, interrupt, or call a tool.
In practice: ChatGPT no longer cuts you off mid-sentence. Ask it to stay quiet and just listen, and it will. And with background noise — passing traffic, nearby conversations — it does a better job of focusing on your voice.
A model that delegates
The second trick: for complex tasks like web search, deeper reasoning, or agentic work, GPT-Live hands off to a frontier model in the background — GPT-5.5 at launch — and keeps the conversation flowing while that work runs. So the voice model doesn’t have to do everything itself; it orchestrates.
Two versions are rolling out globally now: GPT-Live-1 for Go, Plus and Pro users, and GPT-Live-1 mini for the free tier. API access is announced and the waitlist is open. According to OpenAI, more than 150 million people use voice and dictation in ChatGPT every week.
My take
Voice is the front where Anthropic barely competes — Claude simply has no comparable voice mode. That’s exactly why I find GPT-Live relevant for us: it shows where OpenAI is extending a lead that Anthropic deliberately leaves open. Anthropic’s bet is on agentic work on your computer (Cowork, Claude Code), not natural conversation.
I also find the safety section notable. OpenAI expanded its testing to audio-native scenarios — self-harm, psychosis, emotional dependence on AI — and added real-time safeguards, including possible parent notification for teens in distress. That’s the flip side of an AI that feels like a conversation partner: the more natural it sounds, the greater the responsibility.
Sources: