OpenAI launches GPT-Live: AI that listens and talks at the same time

What's new

OpenAI has launched GPT-Live, a new generation of voice models for ChatGPT. The model comes in two versions: GPT-Live-1 and a smaller GPT-Live-1 mini. OpenAI describes it as a new generation of voice models for natural human-AI interaction, now powering ChatGPT Voice. On September 10, 2026, GPT-Live-1 was also made available in OpenAI's API, so companies can build their own voice products on top of the technology.

What's clever

The clever part is that GPT-Live is built with a so-called full-duplex architecture, which means the model can listen and speak at the same time. In practice, this means you can interrupt the AI mid-sentence or add details along the way without the conversation stalling. During conversations, GPT-Live can show that it is following along with sounds like "mhmm" or "yeah," and it can hold a natural pause if you need to think. When a task requires deeper reasoning or a web search, GPT-Live passes the task on to a more powerful model in the background while it continues the conversation. At launch, GPT-5.5 is used as this background model.

Cheaper or better?

For ordinary users, it costs nothing extra. GPT-Live-1 is the default for the paid ChatGPT subscriptions Go, Plus and Pro, while free users get the lighter GPT-Live-1 mini. For developers, the voice layer itself costs $0.05 per minute, while use of the background model for heavy reasoning is paid separately. Compared with Google's competing Gemini 3.8 Live, OpenAI's model is more expensive on paper, but the pricing is structured differently: where Google includes the reasoning itself in the audio model, OpenAI sells the voice layer separately and lets the customer choose and pay for their own background model. On benchmarks, GPT-Live-1 is strong on conversational flow but slightly behind on pure audio-based reasoning compared with Google's model.

What it's good at

GPT-Live is especially suited to tasks where a fluid conversation matters. The review platform Yelp already uses the technology to book restaurant tables by phone, where conversations rarely follow a fixed script and calls need to handle interruptions and details along the way. The language learning service Speak found early on that the technology gave students more time to think before the artificial language teacher responded. The technology is therefore a good fit for customer service, virtual assistants and other situations where you talk back and forth with an AI, rather than just giving it short commands.