ChatGPT stopped making you wait — 150 million get GPT-Live today

When you talked to ChatGPT before, you waited. The model processed your speech, generated a response, and delivered it — then waited for your next turn. That architecture, called turn-based, is how most voice AI has worked since the first commercial deployments. OpenAI has scrapped it.

GPT-Live runs on a full-duplex architecture, which means it processes incoming speech and generates its response at the same time. The practical difference is that you can interrupt the model mid-sentence — and it stops. It can register brief backchannels, questions, or redirections without requiring either participant to complete their turn first. OpenAI began the global rollout today, deploying GPT-Live-1 to paid subscribers (Go, Plus, Pro) and GPT-Live-1 mini to free users, replacing Advanced Voice Mode without an opt-in prompt.

The 150 million people who use ChatGPT voice features weekly will feel the change most clearly in two situations: conversations that shift direction quickly, and the disappearance of the pause that previously signaled the model was finished talking. GPT-Live can signal attention with brief acknowledgements — ‘mhmm’, ‘yeah’ — or simply stay quiet when the speaker is still thinking. When the conversation requires web search, deeper reasoning, or complex work, it delegates silently to a frontier model behind the scenes and returns the result into the live exchange.

The skepticism paragraph matters here. Full-duplex does not mean GPT-Live has become better at understanding what you say — it means it can respond to the rhythm of a conversation more naturally. Early testing suggests that when a speaker delivers a long, dense explanation, the model occasionally acknowledges a statement before it has finished processing the full meaning. The system also runs entirely on OpenAI’s servers, so every voice conversation depends on a low-latency connection that not all of the 150 million weekly users will have in equal measure. Regions with slower infrastructure may find that GPT-Live’s real-time ambitions outpace the network reality beneath them.

There is also a structural question about the migration itself. OpenAI has automatically moved its entire voice-active user base to a new system without requiring consent. The company frames this as an upgrade — and in many respects it is. But it is also a change to the interaction model that millions of people had calibrated their expectations against, delivered without a disclosed retirement date for the previous system.

OpenAI has not announced price changes for current ChatGPT subscribers in connection with GPT-Live. The rollout is expected to reach all users globally within the coming days. Advanced Voice Mode’s official retirement date has not been confirmed.

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