TL;DR: On Oct 16, 2026, OpenAI released prompt-to-voice to all API customers. You describe a voice in text, OpenAI saves it, and you can reuse it in GPT-Live sessions. Creating a voice is free; speech uses normal GPT-Live session billing. These voices work only in GPT-Live, not in the Text to Speech, Realtime or Chat Completions APIs, and each organization can create up to 100 of them.

What happened

OpenAI’s release notes for Oct 16 say it “released prompt-to-voice for all API customers.” You can “create a custom voice from a text description through the API or voice Playground, then reuse it in GPT-Live sessions.”

The same update added nine voices for GPT-Live: Nuri, Juni and Flitz (German); Brise and Sillage (French); Nira (Hindi); Harema and Shida (Japanese); and Noeul (Korean).

OpenAI’s custom voices guide fills in the details:

  • How you create one. Send a request to /v1/audio/voices with type set to prompt, a name and a prompt describing the voice. You don’t need a reference recording or a consent recording.
  • What you get back. A saved voice ID and the original prompt. The response doesn’t include audio or a preview.
  • Optional script. script_hint lets you give up to 2,000 characters of sample speech to guide the voice. If you leave it out, the service writes one from your description.
  • Where it works. “Prompt-generated voices work with GPT-Live. They aren’t supported by the Text to Speech, Realtime, or Chat Completions APIs.” Voices made from a consent recording and sample, which OpenAI limits to eligible customers, still work across those other APIs.
  • The cap. “You can create up to 100 prompt-generated voices per organization. This limit also applies when your application creates voices for its customers.”
  • The cost. “Creating a prompt-generated voice is free. Using it to generate speech in GPT-Live is billed at regular session rates.”

Why it matters

A custom brand voice for a phone or support agent used to mean either picking a stock voice or, for eligible customers, recording consent and sample audio. Now it’s a text prompt and an API call, at no cost to create.

Two limits shape how you’d use it. First, the voice is tied to GPT-Live. If your stack uses the Realtime API or text-to-speech, a prompt-made voice won’t move with you. Second, the 100-voice cap is per organization and includes voices your app creates for its own customers. So a “let every customer design a voice” feature runs out fast.

OpenAI also says voices “aren’t guaranteed to be unique,” so someone else may generate something similar. That matters if you planned to treat a voice as part of your brand.

What to do

  1. Check your stack first. If you’re on GPT-Live, you can use these voices. If you’re on Realtime, TTS or Chat Completions audio, you can’t, so don’t plan around them yet.
  2. Write the prompt like a spec. Describe accent, pitch, tone and pace in plain language, as OpenAI’s example does. Save the prompt with the voice ID; OpenAI says to retain it and reuse the saved voice for consistency.
  3. Create voices on purpose, not per request. Each successful request makes a new saved voice and counts toward the 100 cap. Reuse IDs instead of creating a fresh voice every session.
  4. Budget for the speech, not the voice. Creation is free, but every minute of GPT-Live audio is billed at session rates. Test with real call lengths.
  5. Follow the use rules. OpenAI says to use adult voices only, not to impersonate anyone, and to tell listeners the speech is AI-generated.

Sources & method

Sources

Method. Last checked October 18, 2026. Quotes, limits and pricing details come from OpenAI’s own release notes and API docs. We haven’t tested the feature. No affiliate links on this page. How we work.

Spotted an error? See our corrections policy and tell us.