OpenAI has launched GPT-Live, a new series of voice models for ChatGPT that can listen and speak at the same time, marking the company's most significant overhaul of its conversational voice experience to date. The release uses what OpenAI describes as a full-duplex architecture, allowing users to interrupt, talk over, and converse with the assistant in a way the company says feels like a real phone call.
The launch arrives alongside OpenAI's broader rollout of the GPT-5.6 model family this week, and represents a push to make breaking AI news about voice interfaces feel less like commanding a machine and more like chatting with a person.
What Makes GPT-Live Different
Traditional voice assistants operate in a half-duplex mode: you speak, then you stop, then the system processes your request and responds. Turn-taking is rigid. GPT-Live breaks that pattern with full-duplex audio processing, meaning the model can hear you while it is still talking. That enables natural interruptions, mid-sentence corrections, and overlapping conversation — the kind of back-and-forth humans take for granted but voice AIs have historically struggled with.
According to Reuters, the GPT-Live models listen and speak simultaneously — a capability that has long been a target for the voice AI field. The approach removes the awkward pauses and forced turn-taking that have defined earlier versions of ChatGPT's voice mode.
A notable technical detail: when GPT-Live encounters complex questions, it hands those off to GPT-5.5 in the background. The voice model handles the conversational flow in real time, while a larger reasoning model works on harder tasks behind the scenes — then feeds the answer back. That division of labor, according to OpenAI, drastically improves response quality without sacrificing the responsiveness that makes real-time conversation feel natural.
Availability and Pricing
OpenAI is rolling out the new voice models in two tiers:
- GPT-Live-1 — the full model, available now to paying ChatGPT subscribers (Plus, Pro, and Team plans).
- GPT-Live-1 mini — a smaller, lighter version available to free ChatGPT accounts.
The launch also brings web search directly into the voice conversation. As Search Engine Journal reported, GPT-Live integrates live search into ChatGPT voice, so users can ask about current events or fast-changing information and get answers grounded in fresh web results without switching modes.
A Competitive Voice AI Market
GPT-Live lands in an increasingly crowded voice AI landscape. Google has been pushing its Gemini-powered voice features across Android and its Gemini Live product, while Apple continues to rework Siri around large language models. Anthropic, OpenAI's chief rival, has focused its recent efforts on agentic coding and reasoning models rather than voice.
The full-duplex approach is where the broader industry appears to be heading. By allowing simultaneous listening and speaking, GPT-Live reduces latency and removes the single biggest complaint users have had with voice assistants: the wait. The handoff to GPT-5.5 for complex queries is also a signal that OpenAI views voice not as a stripped-down interface, but as a front door to its most capable models.
Why It Matters
Voice has been treated for years as a secondary, command-driven interface — good for setting timers and checking weather, less useful for nuanced conversation. GPT-Live's bet is that the bottleneck was never the model's intelligence, but the interface: forcing humans to speak in isolated bursts.
If full-duplex conversation works reliably at scale, it changes how people interact with AI on phones, in cars, on smart speakers, and eventually on wearable devices. It also raises familiar concerns — about consent in always-listening devices, about the realism of synthetic voices, and about whether more natural conversation leads users to over-trust an AI's answers.
OpenAI has not detailed specific safety measures for GPT-Live beyond its existing content policies, though the search integration means the model can now pull live information into a voice that may be harder for users to critically evaluate than text they can scroll back through.
What Comes Next
For now, GPT-Live is a ChatGPT feature first and a developer product second. The real test will arrive with API access, when third-party apps, customer-service platforms, and hardware makers can plug full-duplex conversation into their own products. That is also when OpenAI will face pricing pressure from cheaper open-source voice models and from competitors eager to match the feature.
The simultaneous launch with GPT-5.6's public release suggests OpenAI sees voice and reasoning as two halves of the same strategy: a powerful model that thinks, paired with an interface that lets you talk to it like a colleague.
Stay Ahead of AI
For continuous coverage of model releases, voice AI, and everything shaping the industry, follow the latest AI developments at AI Buzz Wire.
Read more AI news →
