GPT-Live finally lands in ChatGPT's work plans, globally

OpenAI shipped GPT-Live-1 on 8 July 2026 to Free, Go, Plus and Pro — and pointedly not to workspaces. That gap just closed: GPT-Live in ChatGPT Voice is now on Edu, Business and Enterprise plans worldwide.

The pitch is full-duplex audio: the model listens and speaks at the same time, so you can cut in mid-sentence and it can cut in on you. It reads intonation, not just words, and hands off anything heavy — web search, deep reasoning — to GPT-5.5 running in the background. It replaces Advanced Voice Mode, which it beat on OpenAI's pleasantness eval (75.5) in 5-10 minute conversations. Business workspaces get two flavours: Voice in Chat for open-ended talk, and Voice in Work and Codex for kicking off tasks and steering agents, with iOS remote access. Desktop voice on macOS and Windows landed 23 July.

The delay was the tell. Consumer voice can ship rough; workspace voice touches admin controls, retention and compliance review, which is exactly why enterprise buyers waited. Now the interesting question is whether anyone actually talks to ChatGPT at a desk full of colleagues — voice UX has always died on the open-plan office, not on the benchmark.

Why it matters: interruptible voice is the first credible shot at agents you steer by talking, not typing — and it's now inside the accounts that pay real money.