OpenAI Posts About GPT-Live Voice Architecture
Official posts outline rebuilt voice stack for continuous interaction at scale.
OpenAI posted that GPT-Live keeps audio flowing continuously so the model can listen while speaking. The rebuilt stack sends real-time audio on a fast path while deeper reasoning and tool calls run asynchronously. Engineers on the voice team shared a technical overview of the client-to-model changes. The company also cut voice session startup from six network round trips to one. The posts link to an OpenAI page describing the realtime system built for responsive ChatGPT Voice conversations.
GPT-Live can listen while it speaks. To make that feel natural at ChatGPT scale, we rebuilt the voice stack from client to model. This new architecture keeps audio flowing continuously, so deeper reasoning and tool use don't interrupt the conversation.
Reactions from ranked influencers
11 postsAudio moves through a dedicated fast path, while deeper reasoning and tool use happen asynchronously. We also reduced voice-session startup from six network round trips to one.
The result is a faster, more natural conversation with ChatGPT Voice from the moment a session starts. How we built it:
GPT-Live was as much a breakthrough engineering project as it was research. Check out this post by @juberti and @zahanm about the global realtime system we built in 6 months.
GPT-Live can listen while it speaks. To make that feel natural at ChatGPT scale, we rebuilt the voice stack from client to model. This new architecture keeps audio flowing continuously, so deeper reasoning and tool use don't interrupt the conversation.
@athyuttamre @juberti @zahanm Is it still using GPT-5.5 (as shown on that diagram) or have you upgraded to 5.6 now?
GPT-Live is a new architecture and stack for realtime audio:
GPT-Live can listen while it speaks. To make that feel natural at ChatGPT scale, we rebuilt the voice stack from client to model. This new architecture keeps audio flowing continuously, so deeper reasoning and tool use don't interrupt the conversation.
Sharing a behind-the-scenes look at the systems powering ChatGPT Voice, by @juberti and I. Very proud of the team that built and shipped this!
GPT-Live can listen while it speaks. To make that feel natural at ChatGPT scale, we rebuilt the voice stack from client to model. This new architecture keeps audio flowing continuously, so deeper reasoning and tool use don't interrupt the conversation.
