GPT-Live brings full-duplex voice to ChatGPT — the pause that gave the AI away is gone

OpenAI replaced its turn-based Advanced Voice Mode with GPT-Live on July 10, 2026, automatically moving 150 million weekly voice users to a simultaneous-processing architecture — no opt-in required, no retirement date announced.

The pause at the end of every ChatGPT voice exchange had a tell. The model finished processing, generated a response, and only then gave you the floor. That rhythm — called turn-based — has defined how most commercial voice AI has worked since its earliest deployments. OpenAI replaced it with GPT-Live.

GPT-Live runs on a full-duplex architecture, processing incoming speech and generating its response at the same time. OpenAI rolled out GPT-Live-1 to paid subscribers on Go, Plus, and Pro tiers, and GPT-Live-1 mini to free users — both on July 10, with no opt-in window, replacing Advanced Voice Mode globally.

The practical shift is the ability to interrupt the model mid-sentence and have it stop. GPT-Live registers brief backchannels, questions, or redirections without requiring either speaker to finish their turn. It can signal attention with a short “mhmm” or “yeah” while the other person is still thinking, or hold silence until the thought lands. When a conversation requires web search, deeper reasoning, or complex work, GPT-Live delegates to a frontier model behind the scenes and returns the result into the live exchange.

Full-duplex architecture does not make GPT-Live better at understanding what is said — it makes it better at responding to the rhythm of how it is said. That distinction matters. Early users reported a pattern the model’s own enthusiasm can create: GPT-Live occasionally acknowledges a statement before it has finished processing the full meaning, particularly during long, dense explanations. Some users found the backchannels themselves over-eager. The system also runs entirely on OpenAI’s servers, making every voice conversation dependent on a low-latency connection that not all 150 million weekly users have equally. Regions with slower infrastructure may find the real-time ambitions outpace the network beneath them.

OpenAI moved its entire voice-active user base to the new system without requesting consent. The company describes GPT-Live as an upgrade — and the full-duplex shift is a real improvement in conversational flow. But it also replaced the interaction model against which millions had calibrated expectations, with no disclosed retirement date for Advanced Voice Mode.

No price changes have been announced for ChatGPT subscribers. An API rollout for developers and enterprises is planned, with early-access sign-ups now open. OpenAI has not confirmed when Advanced Voice Mode will officially retire.

Tags: OpenAI, AI models, chatgpt, voice AI, full-duplex, GPT-Live