OpenAI has launched GPT-Live, a new family of voice models that lets ChatGPT listen and speak at the same time, so a spoken conversation feels less stilted and more like talking to a person. The company rolled it out to ChatGPT users worldwide starting July 8, 2026. During the launch livestream, OpenAI researcher Kundan Kumar described the technology as "one step closer to a truly accessible AGI, a world where talking to AI actually starts to feel like a real conversation." AGI, short for artificial general intelligence, is the industry's loosely defined goal of AI that can match or beat humans across most tasks.
Key Takeaways
- GPT-Live is a new voice model generation that powers an upgraded ChatGPT Voice, rolling out globally from July 8, 2026.
- It uses a full-duplex design, meaning it can listen and speak at the same time, so you can interrupt it naturally.
- For hard questions, it hands the work to a stronger model (GPT-5.5) in the background while keeping the conversation going.
- Two versions ship: GPT-Live-1 for paid Go, Plus, and Pro users, and GPT-Live-1 mini for Free users.
- OpenAI added voice-specific safety features, including crisis support flows and protections for teen users.
What GPT-Live changes about ChatGPT Voice
The headline change is full-duplex audio. Full-duplex means the model can hear you and talk to you at the same moment, the way two people do in a real chat. Older ChatGPT voice modes worked in strict turns: the model waited for you to stop talking, then replied. That created awkward pauses, and a short pause or background noise could be mistaken for the end of your turn, causing the AI to cut in at the wrong time.
With GPT-Live, ChatGPT can decide many times per second whether to speak, keep listening, pause, interrupt, or use a tool. It can drop in small acknowledgements like "mhmm" or "got it" so you know it is following along, and it can stay quiet when you need a moment to think. In one launch demo highlighted by 9to5Mac, a user told the model to interrupt after the tenth letter of the alphabet, and it did.
How the full-duplex model works
OpenAI splits the job into two parts. GPT-Live handles the live back-and-forth of the conversation. When a question needs web search, deeper reasoning, or more complex work, GPT-Live passes that task to a stronger "frontier" model in the background and brings the answer back when it is ready. At launch, that background model is GPT-5.5. OpenAI says it will keep updating which model GPT-Live delegates to as newer ones arrive.
This split lets the voice stay fast and natural while the heavy thinking happens out of sight. You can also pick how hard the model works: Instant for quick replies, or Medium and High when you want it to spend more time reasoning. The faster modes use GPT-5.5 Instant, while Medium and High use GPT-5.5 Thinking.
In OpenAI's own head-to-head human tests, both GPT-Live-1 and GPT-Live-1 mini were strongly preferred over the older Advanced Voice Mode for turn-taking, interruptions, and how natural each chat felt. The company also reported gains on standard benchmarks such as GPQA (expert-level science questions) and BrowseComp (finding hard-to-locate information on the web). SiliconANGLE reported that GPT-Live scored 75.5 on OpenAI's internal pleasantness measure, ahead of the previous voice model.
Smarter answers, better listening, and visual cards
Beyond the conversation feel, OpenAI remastered the nine distinct voices in ChatGPT for GPT-Live. The model is better at focusing on your voice when there is traffic or nearby chatter, and it waits instead of jumping in if you go quiet to gather your thoughts.
ChatGPT Voice can now also show information on screen while you talk. It can display rich visual cards for topics like weather, stocks, and sports, and it still supports search, memory, images, and file uploads during a voice session.
'One step closer to a truly accessible AGI'?
The AGI framing is worth reading carefully. As Gizmodo noted, AGI is used loosely today as both a technical target and a marketing term, and the industry has no single agreed definition. GPT-Live is a big step for natural voice interaction, but it is a conversational layer on top of existing models like GPT-5.5, not a new form of general intelligence on its own. Gizmodo also flagged the privacy question raised by any always-listening assistant, a trade-off that comes with a device designed to hear you continuously.
Safety designed for voice
OpenAI says GPT-Live was built to be safe by default, with extra safety training and new safeguards made specifically for voice. Because voice conversations happen in real time, the system can act while the model is still speaking: it can steer toward a safer answer, surface support resources, or end a higher-risk conversation. For talks involving self-harm, OpenAI adapted ChatGPT's support flows for voice, including expert-vetted crisis helpline support.
The company also added protections for teenagers, training age-appropriate behavior into the model. Parents can decide whether their teen may use ChatGPT Voice through Parental Controls and may be notified in higher-risk situations. Finally, GPT-Live uses a set of preset voices with safeguards so it will not imitate a real person's voice.
Availability and limits
GPT-Live is rolling out now to ChatGPT users worldwide, including in the Philippines, across iOS, Android, and ChatGPT on the web, plus CarPlay support noted by 9to5Mac. GPT-Live-1 becomes the default voice model for Go, Plus, and Pro subscribers, while GPT-Live-1 mini becomes the default for Free users. OpenAI plans to bring the models to its API soon and is collecting developer sign-ups.
There are limits at launch. GPT-Live does not yet support voice with video or screen sharing, so those still use the older Standard and Advanced Voice modes. OpenAI also said the model is tuned for the most popular languages, and for some languages it may have a non-native accent or gaps in fluency.
Trailer
Why It Matters for PH Users
Because GPT-Live is a global ChatGPT rollout, Filipino users get the upgraded voice experience automatically at their plan level, with no separate local launch needed. A more natural, hands-free ChatGPT is useful for commuting, language practice, and quick everyday help. The wider release also lands around OpenAI's broader push of its newer models, following the GPT-5.6 rollout to everyone, and it deepens the local footprint OpenAI already built through the Viber ChatGPT tools for Philippine users. The accent and fluency gaps OpenAI mentioned are worth watching for anyone who speaks Filipino or Taglish to the assistant.
FAQ
Is GPT-Live free to use?
Yes. Free ChatGPT users get GPT-Live-1 mini as the default voice model, while paid Go, Plus, and Pro users get the more capable GPT-Live-1.
Which model does GPT-Live use to answer hard questions?
At launch it delegates complex work, such as web search and deeper reasoning, to GPT-5.5 in the background, then returns the answer while keeping the conversation going.
Can it do video or screen sharing?
Not yet. At launch GPT-Live does not support voice with video or screen sharing. Those features still run on the older Standard and Advanced Voice modes.
Sources:
