Cartesia
4.4
Cartesia builds low-latency speech models: Sonic for text-to-speech, Ink for speech-to-text, and managed voice agents. Free gives 20,000 credits a month and $1 of prepaid agent usage; Pro is $5 a month for 100,000 credits, Startup $49 for 1.25 million and Scale $299 for 8 million, each with matching prepaid agent usage.
- Commercial use OK
- Free tier
- Pricing
- $5/mo
- Free tier
- Free tier
- Commercial use
- Permitted
- Japanese UI
- Not confirmed
- Enterprise contract
- Not confirmed
What it is good for
- Voice applications where response latency is the constraint
- Teams wanting speech synthesis and recognition from one vendor
- Prototyping voice agents cheaply
- Products where the voice is part of the brand
Strengths and trade-offs
Strengths
- Latency is the explicit design target
- Pro at $5 is unusually cheap for a real allowance
- Free tier of 20,000 credits allows genuine testing
- Covers TTS, STT and agents together
Trade-offs
- Smaller voice library than the established players
- Credit consumption per character needs modelling
- Japanese voice quality should be checked directly
Editors' review
Latency is what makes voice feel broken
Users forgive an imperfect voice. They do not forgive a one-second gap before every reply, because it reads as the system not understanding them.
At $5 for the Pro tier, comparing it head to head against your current provider on your own script costs almost nothing.
Verdict
Worth a direct latency comparison. The entry price makes the test trivial.
Compiled from official documentation and public sources, with our own editorial judgement. We do not publish figures we cannot source.