OpenAI's GPT-Live-1 Speaks and Listens at Once
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI launched GPT-Live-1, calling it the company's "smartest voice model" yet at a press briefing where research lead Kundan Kumar said it interrupts users less and waits for them to continue speaking after a pause.
- GPT-Live-1 operates as a full duplex model that can speak and listen at the same time, processing input and output streams continuously — a departure from the older turn-based voice model that sometimes broke conversational flow, per product lead Atty Eleti.
- The model automatically offloads reasoning and web search to text models like GPT-5.5, letting ChatGPT move from researching a topic to discussing findings more quickly.
- ChatGPT Voice gains real-time translation that works while the user is still talking, a command to make the assistant stop speaking until called on again, and acknowledgment phrases like "mhmm," "yeah," and "got it."
- Conversations about weather, stocks, and sports are now supplemented with AI-generated visuals showing forecasts, scores, and other relevant data.
- OpenAI built in safeguards that steer GPT-Live-1 away from harmful responses, terminate chats in "higher-risk" situations, surface "expert-vetted crisis helpline support" during self-harm talks, and produce "age-appropriate" responses for teens — disclosed amid ongoing lawsuits alleging ChatGPT harmed users' mental health.
- GPT-Live-1 is rolling out on iOS, Android, and the web for Go, Plus, and Pro subscribers, while a smaller GPT-Live-1 mini becomes the default for free users.
Why it matters: By shipping a full-duplex voice model that listens while it speaks, OpenAI is closing a long-standing capability gap with rival assistants and turning ChatGPT Voice into a more natural conversational layer rather than a chatbot with a mic. The free tier gets the lightweight GPT-Live-1 mini, broadening access to near-real-time voice, while safety guardrails — including crisis helpline handoffs and teen-appropriate filters — are being framed as core to the rollout rather than bolted on, likely setting the template for how voice AI competitors ship their own safeguards.
