OpenAI's text-to-speech API offers three models (gpt-4o-mini-tts, the newest and most controllable; tts-1, low-latency; tts-1-hd, higher quality) across 13 voices, priced from $12-30 per million characters. Pick it when you want hosted TTS with fine control over tone and delivery via prompting. OpenAI's usage policy requires clearly disclosing to end users that the voice is AI-generated, not human.
Other good fits
Projects already using OpenAI's other APIs that want speech output from the same provider and billing.
+ Pros
Read pros
OpenAI offers three text-to-speech models. gpt-4o-mini-tts is the newest and most capable for realtime applications, letting you prompt for accent, emotional range, intonation, speed and tone rather than just picking a fixed voice and hoping. tts-1 trades quality for lower latency; tts-1-hd trades latency for higher quality. All three draw from 13 built-in voices (alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer, verse, marin, cedar), with OpenAI recommending marin or cedar for the best quality; older models support a smaller subset of these voices.
Pick it when…
You want prompt-level control over how something is said (tone, pace, emotion), not just which built-in voice speaks it.
You're already using OpenAI's other APIs and want speech output under the same account and billing.
Low latency for a realtime application matters more than the highest possible audio quality (use tts-1 for that trade-off).
− Cons
Read cons
The built-in voice set (13 voices) is fixed; there's no voice cloning from a sample the way some other TTS providers offer, so if a project specifically needs a cloned or highly distinctive custom voice, this isn't the right tool. OpenAI's usage policy requires a clear disclosure to end users that the voice they're hearing is AI-generated, not human; this is a real compliance requirement, not an optional nicety, and needs to be visible wherever the audio is heard.
Pick something else when…
You need voice cloning from a sample.XTTS-v2 is built specifically for that, with its own consent and licensing considerations.
You want a free, self-hosted model instead of a metered API.Kokoro runs on your own hardware with no per-character billing.
You need this integrated into a broader realtime conversational pipeline, not just one-off TTS.OpenAI Realtime handles speech-to-speech conversation directly rather than text-to-speech alone.
Avoid if
You can't or won't add a clear AI-generated-voice disclosure; OpenAI's usage policy requires it.
Context
gpt-4o-mini-tts is the model to reach for by default for new projects: it's both the newest and the one with the most controllable delivery. tts-1 and tts-1-hd remain available for their specific latency/quality trade-off, priced per character rather than per token.
Requirements
This is a hosted API call; you need an OpenAI API key and network access, not local hardware. Voice availability varies by model, so confirm your chosen voice is supported by the specific model you're calling before building a workflow around it.
There's no local hardware cost since this is a hosted API. Running effort is mostly watching character volume for cost, and building the disclosure requirement into your product from the start rather than retrofitting it later.
First task
Documentation-based suggestion — not run here.
Generate the same line with two different tone prompts.
Get API access and pick gpt-4o-mini-tts with the marin or cedar voice.
Write one short sentence and generate it twice: once with a prompt asking for a calm, informative tone, and once asking for an upbeat, enthusiastic tone.
Listen to both and confirm the tone difference is actually audible, not just requested.
Add a clear disclosure (in your app or alongside the audio) that this is an AI-generated voice.
Done when you can hear a clear difference between the two tone prompts, and you've written the disclosure text you'll actually ship with the audio.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on OpenAI's text-to-speech guide and pricing documentation (developers.openai.com, redirected from platform.openai.com/docs), checked 23 September 2026. No API key was used and no audio was generated for this review.
Turn written content into spoken narration with OpenAI TTS models.
Add speech output to chatbots and accessibility experiences.
Pair text generation and transcription within OpenAI workflows.