Voice-agent developers who need very fast speech and quick voice cloning; a free monthly allowance makes it cheap to trial.

Kenning's take

What it does

Sonic 3 is Cartesia's real-time text-to-speech model, built for voice agents, with first audio in under 90 milliseconds. It picks up emotion from the text, supports tags such as laughter, clones a voice from about 10 seconds of audio and works in many languages, with the current Sonic-3.6 covering 44. Developers use it through Cartesia's API on free and paid plans.

  • First audio in under 90ms
  • Emotion read from the transcript and [laughter] tags
  • Instant voice cloning from 10 seconds of audio
  • 44 languages in Sonic-3.6
  • Custom pronunciation dictionaries

Who it suits. Developers building voice agents, phone systems and narrated apps.

Key facts

Category
AI models
Made by
Cartesia
Launched
2025
How to get it
through the Cartesia API and playground; also on Amazon SageMaker JumpStart
Works on
API, Web
Free plan
20K credits a month on the Free plan
API price
credit-based; Pro's 100K credits give about 133 minutes of text-to-speech a month

Pricing

PlanPriceBilled
Pro$5month
Startup$49month
Scale$299month

source

FAQ

How fast is Sonic 3?

Cartesia says first audio arrives in under 90 milliseconds.

Is Sonic 3 free?

There is a free plan with 20,000 credits a month; paid plans start at $5 a month.

Can Sonic 3 clone a voice?

Yes, from about 10 seconds of audio.

Facts from the maker's own pages and the sources listed. Checked 2026-10-11.