OpenAI Launches GPT-Live, a Voice Model for Real-Time Conversations

OpenAI has rolled out GPT-Live, a new generation of voice models in ChatGPT designed to make talking with the AI assistant feel more like talking to another person. The models are rolling out globally on iOS, Android, and the web app, replacing the previous Advanced Voice Mode.

How Full Duplex Works

GPT-Live is built on a full-duplex architecture, meaning the model can listen and speak simultaneously instead of waiting for the user to finish a sentence. That lets it drop in short acknowledgments like mhm or yeah, naturally match the rhythm of a conversation, and stay quiet when the other person needs a moment to think.

This marks a break from the previous three-step pipeline of speech to text, processing by a language model, and text back to speech, a chain that introduced noticeable lag and made it hard to interrupt naturally.

Two Variants for Different Plans

OpenAI introduced two versions of the model. GPT-Live-1 mini has replaced Advanced Voice Mode by default for all ChatGPT users, including those on the free plan. The larger GPT-Live-1 is now the default for paid Go, Plus, and Pro subscriptions.

Over time, we think this will unlock the ability to use voice as a primary interface to the computer - Atty Eleti, head of ChatGPT voice product at OpenAI

Voice as an Interface

Behind the scenes, GPT-Live relies on the GPT-5.5 model for information lookup, deeper reasoning, and tasks that require multiple steps. When a conversation calls for it, the model hands the query off to the latest frontier model and returns with an answer without breaking the natural flow of the conversation. OpenAI says GPT-Live will keep drawing on future frontier model releases as they launch.

The model can also present information visually alongside its spoken answer, which matters for things like numerical data, directions, or product comparisons.

Competition in Voice Conversations

The launch of GPT-Live fits into a broader race among tech companies toward more natural voice interfaces. Apple and Amazon are pursuing similar directions with their own assistants, while the startup Sesame is promoting its own conversational AI assistant.

For users and businesses in Poland, this means ChatGPT's voice experience will noticeably smooth out in the coming weeks, which could speed up adoption of voice-based customer service bots and mobile app assistants built on OpenAI's API.

Sources: TechCrunch (techcrunch.com), Technology.org (technology.org), SiliconANGLE (siliconangle.com)