OpenAI Launches GPT-Live: Full-Duplex Voice That Listens and Talks at the Same Time

TL;DR: OpenAI shipped GPT-Live on July 8 — GPT-Live-1 and GPT-Live-1 mini replace Advanced Voice Mode with a full-duplex architecture that listens and speaks simultaneously. API access is gated; gpt-realtime-2.1 stays the developer path today.

OpenAI on July 8 launched GPT-Live: two voice models (GPT-Live-1 for paid ChatGPT tiers, GPT-Live-1 mini for free) that replace the old turn-based Advanced Voice Mode. The architecture is genuinely different: full-duplex, meaning the model processes incoming audio and generates speech at the same time. It can produce filler words ("mhmm", "yeah"), handle interruptions naturally, and delegate to GPT-5.5 in the background for complex reasoning while continuing to speak. Rolling out now on iOS, Android, and ChatGPT.com.

API access is not yet available — developers can sign up at openai.com for notifications. A System Card was released the same day covering safety evaluations.

Why it matters: GPT-Live is a ChatGPT product, not an API you can call today. If you're building voice-agent products, gpt-realtime-2.1 (already shipping in the API) remains your production path. Keep an eye on the API release; when full-duplex voice arrives for developers, it changes what's possible for ambient and conversational agents.

Sources: Introducing GPT-Live (OpenAI) · VentureBeat: GPT-Live full-duplex launch · 9to5Mac: ChatGPT voice upgrade with GPT-Live