OpenAI ships GPT-Live 1 voice model with delegations

This tutorial details how developers can build and test full-duplex voice agents using OpenAI's GPT-Live 1 model and WebSocket API. The author explains how GPT-Live handles conversations concurrently by delegating complex reasoning, web searches, or custom function calls to a backend text model. Practical examples include connecting via Node.js, streaming audio data, configuring tools, and handling interruptions. Open article →