OpenAI Launches GPT-Live, A Full-Duplex Voice Model For Real Conversations

OpenAI has launched GPT-Live, a new generation of voice models built on a full-duplex architecture that lets ChatGPT listen and speak at the same time. The company announced the release on July 8, 2026, and immediately rolled it out as the new default for ChatGPT Voice across iOS, Android, and ChatGPT.com.

Continuous Listening, No More Turn-Taking Gaps

GPT-Live continuously processes audio while generating output, letting the model decide many times per second whether to speak, keep listening, pause, interrupt, or invoke a tool. In practice, ChatGPT can now emit acknowledgements like "mhmm" or "got it", hold silence while a user thinks, and pick up a translation mid-sentence — behaviours previous cascaded and turn-based voice stacks could not deliver.

GPT-5.5 In The Background For The Hard Work

The company also decoupled conversational voice from deeper reasoning. GPT-Live handles the live audio channel and delegates web search, agentic tasks, or heavy reasoning to GPT-5.5 in the background, keeping the conversation flowing while a second model works. Users can pick Instant, Medium, or High reasoning modes; GPT-Live-1 becomes the default for Go, Plus, and Pro subscribers, while GPT-Live-1 mini becomes the default on the free tier.

Visual Answers, Safer Voice, API Next

ChatGPT Voice now overlays rich visual cards for weather, stocks, sports, and maps while GPT-Live keeps talking, and can lean on the assistant's memory, search, and file uploads. OpenAI says GPT-Live was red-teamed on audio-native evaluations covering self-harm, psychosis, emotional reliance, violence, and sexual content, and shipped built-in safeguards that can steer or end a call in higher-risk situations, plus parental controls for teens. Video and screen-sharing are not yet supported. Developers can already sign up for API access.

The Voice-First Push Continues

The launch caps a busy stretch for OpenAI, coming a day before the Microsoft 365 Copilot switch to GPT-5.6 and following the debut of the Broadcom-built Jalapeño inference chip and ChatGPT Work. It also lands as rivals push physical AI stacks like Mistral's Robostral Navigate deeper into robotics.

Reporting based on OpenAI's official announcement and coverage from TechCrunch.