OpenAI GPT-Live arrived on July 8, 2026, and it changes what talking to ChatGPT actually feels like. Instead of waiting for you to finish a sentence before it responds, this new release listens and speaks at the same time, so a ChatGPT Voice conversation finally moves like a conversation with a person instead of a walkie-talkie exchange. GPT-Live ships as two models, and OpenAI has already made it the default voice experience for ChatGPT on iOS, Android, and the web. OpenAI describes this as the biggest change to ChatGPT Voice since it first launched, and if ChatGPT Voice has sounded less robotic to you this week, that is OpenAI GPT-Live doing its job. If you were searching for the exact OpenAI GPT-Live release date, OpenAI confirmed the rollout began on July 8, 2026, not on some future date. You can read the full details in OpenAI's official announcement.
The core idea behind OpenAI GPT-Live is a full-duplex voice architecture. In plain terms, it can process what you are saying and generate its own speech at the exact same time, rather than taking strict turns the way ChatGPT Voice used to. Advanced Voice Mode, the feature this release replaces, still worked in discrete turns behind the scenes: it had to detect silence before deciding you were done talking, and any pause, cough, or noisy room could trick Advanced Voice Mode into jumping in at the wrong moment. OpenAI built GPT-Live to remove that guesswork by continuously listening while it talks, which is exactly why OpenAI calls it a full-duplex voice model rather than just a faster version of Advanced Voice Mode.
The headline OpenAI GPT-Live features are the full-duplex architecture itself, the split between GPT-Live-1 and GPT-Live-1 mini, and the sub-200ms response speed, and together those OpenAI GPT-Live features are what people mean when they ask what is OpenAI GPT-Live in the first place. This distinction matters more than it sounds on paper. A phone call between two humans is full-duplex by nature, since both people can technically talk over each other, pause, and pick the thread back up without confusion. OpenAI designed GPT-Live to bring that same rhythm to ChatGPT Voice. Where Advanced Voice Mode treated every exchange like a radio handset, this release treats a conversation the way people actually have one, with natural overlap, quick check-ins, and comfortable full-duplex silence when you are still thinking through what to say next. OpenAI says the shift away from Advanced Voice Mode toward a full-duplex voice design is the single biggest architectural change ChatGPT Voice has seen since it launched. Put simply, when you weigh OpenAI GPT-Live vs Advanced Voice Mode, the older system waited its turn and the new one does not.
OpenAI GPT-Live is not a single model but a small family. The larger GPT-Live-1 model now powers ChatGPT Voice by default for Go, Plus, and Pro subscribers, while GPT-Live-1 mini has become the default for Free tier accounts, taking over the exact spot Advanced Voice Mode used to occupy. Both versions are rolling out globally, not region by region, across the iOS app, the Android app, and ChatGPT.com in the browser, and OpenAI has confirmed there is no waitlist for either one.
For anyone who just wants to know how to turn on the OpenAI GPT-Live voice mode, the short answer is that you do not have to, since it is now the automatic default. For anyone tracking what is new in ChatGPT Voice, this rollout is worth watching closely because it touches every paid tier plus the free tier at once, which is a broader launch than most ChatGPT Voice updates get from OpenAI. If your account defaults to Advanced Voice Mode today, expect ChatGPT Voice to switch over to OpenAI GPT-Live automatically once OpenAI's update reaches you, since it is meant to fully replace Advanced Voice Mode rather than sit alongside it as an option. OpenAI has not said whether Advanced Voice Mode will return once the new release finishes rolling out everywhere.
Because GPT-Live processes audio continuously instead of waiting for a full turn, the model has to make a lot of small decisions every second: should it keep listening, start speaking, pause, interrupt, or hand a request off to a tool. OpenAI built it to make these interaction decisions many times per second, which is what lets ChatGPT Voice react the way a human listener would. During a call, OpenAI GPT-Live can drop in a quick "mhmm" or "yeah" to show it is following along, jump in with a fast clarifying question, or simply stay quiet while you are gathering your thoughts instead of assuming you are finished.
This turn-taking behavior is the biggest usability change OpenAI GPT-Live brings to ChatGPT Voice. Early users have noted that it sometimes leans a little too hard into these backchannel cues, with the frequent "mhmm" and "yeah" feeling more distracting than reassuring in some conversations. OpenAI expected some of that feedback, since a launch built around full-duplex voice means the model is making live judgment calls about when to speak instead of following a fixed script, and OpenAI has already indicated it is tuning GPT-Live based on how people actually use ChatGPT Voice day to day.
Speed is a big part of why OpenAI GPT-Live feels different in practice. The previous turn-based Realtime pipeline behind ChatGPT Voice carried roughly 1,700 milliseconds of delay before the first word of a reply, and OpenAI says that gap was long enough to feel like an awkward pause in a normal chat. This release cuts that down to sub-200ms latency for most exchanges, which is fast enough that a response from OpenAI GPT-Live feels immediate rather than computed.
Reaching that speed without losing depth required a design choice from OpenAI: GPT-Live handles the flow of conversation itself, but when a question needs real reasoning, a web search, or a longer lookup, it quietly delegates that work to GPT-5.5 running in the background. GPT-Live keeps talking, filling the gap naturally with ChatGPT Voice still active, and then folds the GPT-5.5 result back into the conversation once it is ready instead of leaving you staring at a spinner. That combination is what lets OpenAI GPT-Live stay fast for small talk while OpenAI still gives it access to deeper GPT-5.5 reasoning for harder questions, all inside one continuous voice session.
Since OpenAI GPT-Live is a ChatGPT product feature rather than a public API right now, there is no code to write to use it. Getting to it only takes a few steps:
Because OpenAI GPT-Live replaced Advanced Voice Mode as the default rather than launching as an opt-in toggle, most people will not need to change any settings at all to start using it; ChatGPT Voice switches over automatically as OpenAI's update reaches each account.
Developers hoping to build with OpenAI GPT-Live directly will need to wait a little longer. OpenAI has confirmed that API access is coming, but no specific date has been announced yet, and interested developers can currently only sign up through OpenAI's notification form to hear when access opens. This is separate from OpenAI's existing Realtime API used for real-time voice agents, which already supports remote MCP servers, image inputs, and SIP-based phone calling for production voice agents, though that Realtime API still runs on the older turn-based pattern that OpenAI GPT-Live is designed to move past.
Reaction to OpenAI GPT-Live in its first days has been largely positive, with most users saying full-duplex voice genuinely feels more natural in ChatGPT Voice than Advanced Voice Mode ever did. The main complaint circling around it so far is that "over-enthusiasm" issue with backchannel filler words showing up too often, and a smaller number of early reports describe GPT-Live interrupting or reacting mid-sentence in ways that felt premature, though OpenAI appears to be tuning that quickly. For anyone who spends a lot of time talking to ChatGPT Voice instead of typing, OpenAI GPT-Live is the biggest change to that experience since Advanced Voice Mode first launched, and OpenAI has said it sets the direction for where real-time voice agents across the industry are likely headed next.