Full-duplex voice has arrived in ChatGPT. OpenAI’s release notes confirm that both GPT-Live-1 and GPT-Live-1 mini can listen and speak at the same time, making turn-taking and interruptions feel natural in a way that push-to-talk or sequential-response voice systems don’t. The rollout covers chatgpt.com, iOS, and Android for consumer plans, Free, Plus, and Pro.
The feature split is straightforward. Paid subscribers, Plus and Pro, default to GPT-Live-1. Free users default to GPT-Live-1 mini. Both tiers get the same core full-duplex architecture; the mini variant is optimized for the performance constraints of the free tier. OpenAI hasn’t disclosed pricing differences between the two, since both are bundled into existing subscription plans rather than billed separately.
Why it matters
The catch is the enterprise wall. ChatGPT Business, Enterprise, and Edu workspaces won’t see GPT-Live-1 at launch, and OpenAI hasn’t said when they will. For organizations that have been evaluating voice AI for customer-facing or internal workflow applications, this is a meaningful gap. Consumer users get the full-duplex experience today; enterprise teams are watching from the sideline.
Who This Affects
GPT-Live-1 also doesn’t replace Advanced Voice Mode, it works alongside it. Video and screen sharing remain in Advanced Voice Mode, which continues to be available for eligible subscribers. The two systems cover different use cases: GPT-Live-1 handles conversational voice integrated with the chat interface, web search, memory, and visual widgets; Advanced Voice Mode handles the richer multimedia interactions. That division of labor is deliberate, and it tells you something about where OpenAI thinks voice AI is mature enough to ship broadly versus where it still needs a controlled environment.
Context
OpenAI published a GPT-Live system card on July 8 alongside the launch, confirmed via . According to OpenAI’s published system card, the model includes a voice-specific safety architecture capable of real-time input and output checking, with spoken safety responses for high-risk conversations. The system card’s existence signals that OpenAI is treating voice modality as a distinct safety surface, not simply porting the text-safety stack to audio. That framing matters for anyone evaluating voice AI for regulated industries or sensitive use cases.
What to watch
Three things are worth tracking. First, when does GPT-Live-1 reach Business and Enterprise workspaces, OpenAI flagged “at launch” as the exclusion window, which implies a future availability date that hasn’t been announced. Second, what does the API rollout look like, developer access is currently waitlisted, and the API pricing structure for GPT-Live-1 hasn’t been disclosed. Third, whether Advanced Voice Mode gets deprecated or maintained as a permanent parallel offering as GPT-Live-1 matures.
TJS synthesis
Don’t treat this as a consumer-only story. The enterprise exclusion is the signal worth reading. OpenAI is shipping full-duplex voice to tens of millions of consumer users before it’s available in any managed workspace, a sequencing choice that reflects where the safety architecture is and isn’t ready. If you’re on an enterprise or education plan and building voice AI use cases, the timeline remains open. If you’re a developer waiting for API access, get on the waitlist now, usage data from the consumer rollout will shape what gets released to the API and when.
Sources: Reuters, VentureBeat, OpenAI Help Center.