OpenAI ships GPT-Live full-duplex voice for ChatGPT

Category: AI
Status: confirmed
Sources: OpenAI announcement, Simon Willison, HN discussion
Summary: OpenAI launched GPT-Live on 2026-07-08, a full-duplex voice model that listens and speaks at the same time rather than processing separate turns. OpenAI says the model makes interaction decisions many times per second (speak, keep listening, pause, interrupt, or invoke a tool) and delegates questions that need web search or deeper reasoning to a frontier model behind the scenes, using GPT-5.5 at launch. Two variants rolled out to ChatGPT users: GPT-Live-1 as the default voice model for Go, Plus, and Pro, and GPT-Live-1 mini for free accounts. API access is stated to follow.
Comments: HN commenters report the full-duplex handling fixes the prior voice mode's habit of cutting off when the user makes a sound, and note the demo included live speech translation.
Why it matters: Full-duplex voice with backend model delegation is a concrete architecture shift for voice agents, and the stated API availability would bring the pattern to developers.
Follow-up: Track the API release, latency and pricing when it ships, and independent evaluation of the delegation behavior.