On Oct 16 OpenAI made prompt-to-voice generally available to all API customers. Developers describe accent, pitch, timbre and pace in plain text and create a saved custom voice through /v1/audio/voices or the voice Playground, then reuse it in GPT-Live sessions. Creating a voice is free; speech uses regular GPT-Live session billing, with up to 100 prompt-generated voices per organization. OpenAI also added nine natural GPT-Live voices in German, French, Hindi, Japanese and Korean.
Key Takeaways
- ✓Generally available to all API customers from Oct 16; no reference or consent recording needed
- ✓Creating a voice is free; speaking in GPT-Live and previewing in the Playground use regular session billing (per OpenAI)
- ✓Up to 100 prompt-generated voices per organization; optional script_hint of up to 2,000 characters
- ✓Nine new natural voices: 3 German, 2 French, 2 Japanese, 1 Hindi and 1 Korean
- ✓Limits: prompt voices work only with GPT-Live, not TTS, Realtime or Chat Completions; voice creation is not Zero Data Retention compatible
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Core Background & Industry Pain Points
Voice is often the hardest part of shipping a voice agent: preset voices sound generic, and a branded voice usually means hiring a voice actor, recording samples and handling consent. OpenAI opened the GPT-Live-1 API in September, and on Oct 15 brought GPT-Live-1.1 to ChatGPT, Codex and dots, saying 1.1 is coming soon to the API. This release fills the customization gap: describe a voice in text and get one.
Architecture Highlights & Mechanism
Send type: "prompt", a name and a description to /v1/audio/voices, covering accent, perceived age, pitch, timbre and pace. model defaults to auto or can be pinned to 2026-10-08; an optional script_hint supplies up to 2,000 characters of sample speech. The response returns a voice ID that you set in session.audio.output.voice.id, and OpenAI recommends repeating the original description in session.instructions. Voices are chosen at session start; switching requires a new session.
Benchmarks & Measured Performance
OpenAI has published no similarity or naturalness benchmarks. Official limits: creation is free, speech uses regular GPT-Live session billing, and each organization can save up to 100 prompt voices. Voices cloned from recordings remain limited to eligible customers, capped at 20 per organization with samples of 30 seconds or less. The nine new natural voices are Nuri, Juni and Flitz (German), Brise and Sillage (French), Nira (Hindi), Harema and Shida (Japanese), and Noeul (Korean).
Developer Quick Start
Compare a few descriptions in the voice Playground using the same passage, changing one quality at a time, then save and reuse the winning ID. Prompt voices do not work with TTS, Realtime or Chat Completions, and voice creation is not available under Zero Data Retention, though GPT-Live speech itself is ZDR-eligible. OpenAI bans imitating real people or characters and children's voices, and asks you to disclose that speech is AI-generated.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.