On October 16, 2026, OpenAI released prompt-to-voice for all API customers. Developers can describe a voice through the API or voice Playground and reuse the result in GPT-Live sessions.
Design from a description
This feature uses text to design a voice and should not be described as cloning someone from a recording. Start with desired pace, tone, and use case, then listen to the result. A description is not a guarantee of exact control over every nuance.
Creation versus speech billing
Creating the voice is free; speech generation uses normal GPT-Live session billing. Nine natural voices in German, French, Hindi, Japanese, and Korean were also added. The API’s voice list should not be assumed to match ChatGPT Voice’s consumer menu.
Test with a fixed script
Compare dates, numbers, names, and long-sentence pauses using the same script before release. The same day’s Agents API artifact tools and Ultrafast additions are separate capabilities; they do not change the billing boundary for generated speech.
