What speech actually costs per character
List prices for text to speech vary by 40×. Here is what is behind the number, and how to compare providers without getting fooled by credits.
Every text to speech provider publishes a different unit. Some sell credits, some sell minutes, some sell characters, and some bury the rate inside a subscription tier. The first job when comparing them is to put everything on the same denominator.
We use characters of input text. It is the only unit that maps cleanly to what you send over the wire, and it does not change when a voice speaks faster or slower. As a rule of thumb, 900 characters is about one minute of speech at a normal pace.
The published rates
Across the major hosted providers, list prices per 1,000 characters currently run from roughly half a cent to ten cents. That is a 20× spread for output that, to most listeners, is difficult to tell apart in a product context.
- Self-hosted open models: around $0.006, before you pay for a GPU or an on-call engineer.
- OpenAI and Azure: $0.015–$0.016 for standard neural voices.
- Google and Deepgram: around $0.030.
- ElevenLabs: $0.040 on Flash, $0.080 on the flagship model.
- MiniMax: $0.060 turbo, $0.100 for the HD model.
Why the spread exists
The expensive end of that list is not only inference. It is research, voice cloning, music and sound-effect models, per-seat sales teams, and the margin required to fund all of it. You pay for the whole lab even if you only need the speech endpoint.
The cheap end is usually a narrower product: fewer voices, fewer languages, or a self-hosted model that shifts operational cost onto you. The honest comparison is total cost of ownership, not the sticker rate.
How to compare without a spreadsheet
Take a real paragraph from your product, count its characters, and multiply by the list rate. Then add the monthly minimum, any per-seat fee, and the cost of the engineering time you will spend on retries and audio post-processing. That total is the number worth comparing.
For most products under a few million characters a month, the difference between a half-cent and a cent and a half per thousand characters is the difference between a rounding error and a line item. The bigger risk is picking a provider whose pricing model changes underneath you.
Try it with your own text.
322 voices across 75 languages, plus transcription, on one API key. Free tier included.
Start free