OpenAI lets developers design custom GPT-Live voices from a text prompt

OpenAI has opened prompt-to-voice to every API customer, according to its API changelog entry for October 16, 2026. Developers can now describe the voice they want in plain text, either through the API or in the voice Playground, and reuse the result in GPT-Live sessions. GPT-Live is OpenAI's full-duplex voice model, meaning it can keep talking and listening while a backend model or agent handles the reasoning and tool calls.

Creating a prompt-generated voice is free. Speech produced with it is billed at the normal GPT-Live session rate, which OpenAI lists at $0.05 per minute, charged by the second, with backend model and tool use billed separately. The same update added nine natural-sounding voices for German, French, Hindi, Japanese and Korean.

A day earlier, on October 15, OpenAI added a native link between GPT-Live and its Decisions API, which turns text and images into typed answers quickly using the small GPT-6 Luna model. With that link, a voice app can make decisions while the user is still speaking, or react to what the voice model says, either by letting Live hand work to Decisions directly or by triggering decisions as transcript pieces arrive.

Why it matters

Voice agents no longer need a stock voice or a separate text-to-speech vendor, which makes branded, multilingual voice products cheaper to build.

Source: OpenAI