OpenAI launches GPT-Live: full-duplex voice models for ChatGPT with live translation and human-like turn-taking
OpenAI on July 8 unveiled GPT-Live, a pair of full-duplex voice models that can listen and speak at the same time, turning ChatGPT into a more human conversationalist.
Full-duplex architecture
The core of GPT-Live is full-duplex: it processes audio continuously while generating speech, allowing interruptions, interjections, and smoother turn-taking. OpenAI’s research blog noted “the model can therefore make interaction decisions many times per second: whether to speak, continue listening, pause, interrupt, or invoke a tool.” This addresses the brittleness of earlier voice modes, which relied on silence gaps and often interrupted at the wrong moment.
You can even forget you’re talking to an AI. How should I put it? It’s a really, really amazing feeling.
Background delegation to GPT-5.5
When a query requires deeper reasoning or web search, GPT-Live hands it off to GPT-5.5 behind the scenes. The voice model keeps the conversation going with filler phrases, then seamlessly integrates the answer. Atty Eleti, voice product lead, said the design mirrors human dialogue:
We keep the conversation going while we think in the background.
The model also surfaces visual cards for weather, stocks, and sports, and users can set reasoning levels to Instant, Medium, or High.
Real-time translation and silent listening
During a demo, research engineer Yuchen Zhang spoke in Chinese as GPT-Live-1 translated his words into English in real time, one of the features OpenAI hopes will broaden its appeal. The model can also stay silent for extended periods, absorbing context until addressed, and answers to a wake phrase, "Hey Chat" in the demo.
OpenAI says GPT-Live-1 and its smaller sibling outperformed the previous Advanced Voice Mode in 5–10 minute test conversations and made big gains on scientific reasoning and agentic web-search benchmarks.
Rollout and ambitions
GPT-Live-1 becomes the default voice for ChatGPT’s Go, Plus, and Pro tiers, while free users get GPT-Live-1 mini. The models are rolling out across iOS, Android, and the web starting today, replacing Advanced Voice Mode. Over 150 million people already use ChatGPT’s Voice and Dictation weekly, the company said, and Eleti described voice as the future primary interface for agentic computing:
Over time, we think this will also unlock the ability to use voice as a kind of primary interface to computing, and to manage increasingly complex long-running agentic work.
