GPT-Live is OpenAI’s new full-duplex voice model that replaces the previous ChatGPT Voice feature, launched July 8, 2026. Unlike the old turn-based system, GPT-Live listens and speaks simultaneously — acknowledging you with “mhmm” and “got it” mid-sentence, interrupting when needed, and routing complex tasks to GPT-5.5 in the background without dropping the conversation. GPT-Live-1 is now the default for paid ChatGPT users.
- Launch date: July 8, 2026
- Full-duplex: AI listens and speaks simultaneously — no more robotic turn-taking
- Natural acknowledgments: says “mhmm” and “got it” while you’re still talking
- Smart delegation: routes complex queries to GPT-5.5, returns answers seamlessly
- Plans: GPT-Live-1 for paid users (Go, Plus, Pro); GPT-Live-1 mini for Free tier
- Not yet available: video and screen sharing — OpenAI says it’s coming
What GPT-Live Actually Does Differently
The core problem with previous AI voice — including ChatGPT’s original voice mode — was the architecture. It was turn-based: you finished speaking, the AI processed your input, then responded. That rhythm felt robotic because real human conversation doesn’t work that way. People say “yeah, exactly” or “wait, back up” while the other person is still mid-sentence.
GPT-Live is full-duplex, meaning the model is simultaneously listening and speaking. According to OpenAI via MacRumors, the model makes decisions on whether to speak, listen, pause, interrupt, or use a tool “multiple times per second.” That decision frequency is what makes the interaction feel natural — the AI is continuously attending to the conversation rather than waiting for a silence to begin its response.
OpenAI describes the change directly: “Talking with ChatGPT should now feel much more like a real conversation. You can interrupt with a question, pause to gather your thoughts, or ask ChatGPT to slow down. It naturally acknowledges what you’re saying with phrases like ‘mhmm’ or ‘got it,’ so you know it’s following along. We’ve also remastered the nine distinct voices in ChatGPT for GPT-Live.”
The Delegation Architecture: GPT-Live-1 and GPT-5.5 Together
GPT-Live-1 is a lightweight real-time voice model — optimized for low latency, not maximum reasoning depth. When a question requires web search, complex math, or deep analysis, GPT-Live-1 hands the task off to GPT-5.5 (OpenAI’s current frontier model) in the background, then brings the result back into the voice conversation without interrupting the flow. The user never has to wait in silence while the bigger model thinks — the voice model keeps the conversation going.
This is conceptually similar to how Apple’s Siri routes hard questions to ChatGPT or Claude — a latency-optimized front end paired with a more capable backend. The difference is that GPT-Live manages the handoff itself, invisibly, within a single continuous voice session.
Additional capabilities in GPT-Live: live translation, and visual cards displayed during conversation for weather, stocks, and sports. You access it via the Voice button in ChatGPT on any paid plan.
How GPT-Live Compares to Google Gemini Live
GPT-Live is not the first full-duplex AI voice product. Google launched Gemini Live in 2025 with similar real-time voice capabilities, and Amazon added real-time conversation to Alexa+ on Echo devices. OpenAI is positioning GPT-Live as the most natural-feeling of these — the “mhmm” acknowledgment feature directly addresses a criticism that AI voice conversations felt cold and one-sided even when technically functional.
What differentiates GPT-Live’s framing: the delegation to GPT-5.5. Gemini Live also routes complex tasks to Gemini’s reasoning models, but OpenAI is being explicit about the architecture — GPT-Live-1 is a new model class, not a capability toggle on an existing model. Whether the real-time feel lives up to the positioning will depend on latency in practice, which early users are still assessing.
For a broader view of where OpenAI’s product strategy is heading, see OpenAI’s new Lockdown Mode, which rolled out the same week and points to a company increasingly focused on trust and control features alongside new capabilities.
What’s Still Missing
GPT-Live does not yet support video or screen sharing. This is a notable gap: a preview of GPT-4o’s Advanced Voice Mode showed video integration, and that feature has been among the most anticipated. OpenAI confirmed video and screen sharing support is coming but gave no timeline. Until that arrives, GPT-Live remains a voice-only experience — powerful for audio, but not yet the multimodal agent some had anticipated from the initial GPT-4o demo.
Frequently Asked Questions
What is GPT-Live?
GPT-Live is OpenAI’s new full-duplex voice model launched on July 8, 2026, that replaces the previous ChatGPT Voice feature. It allows ChatGPT to listen and speak simultaneously, making conversations feel more natural than the previous turn-based system. GPT-Live-1 is the default model for paid ChatGPT users; GPT-Live-1 mini serves the Free tier.
How is GPT-Live different from the old ChatGPT Voice?
The previous ChatGPT Voice was turn-based: you spoke, the AI processed your words, then responded — a back-and-forth that felt robotic. GPT-Live is full-duplex, meaning the model listens and speaks at the same time. It can acknowledge you mid-sentence with phrases like “mhmm” or “got it,” interrupt when appropriate, and continue naturally without the awkward pauses that characterized earlier voice AI.
Which ChatGPT plans get GPT-Live?
GPT-Live-1 is the default voice model for paid ChatGPT plans — Go, Plus, and Pro. Free-tier users get GPT-Live-1 mini, which is a lighter version of the same full-duplex architecture. All users access GPT-Live via the Voice button in the ChatGPT app.
Does GPT-Live support video or screen sharing?
Not yet. As of the July 8, 2026 launch, GPT-Live is voice-only — it does not support video or screen sharing. OpenAI confirmed these features are coming but provided no specific timeline. Video integration has been one of the most anticipated features since OpenAI’s GPT-4o Advanced Voice Mode preview.
How does GPT-Live handle complex questions?
GPT-Live-1 is optimized for low-latency real-time voice but is not OpenAI’s most powerful reasoning model. When a question requires web search, deep analysis, or complex computation, GPT-Live-1 delegates the task to GPT-5.5 in the background and brings the result back into the conversation without interrupting the voice session. The handoff is invisible to the user.
