Cartesia (cartesia.ai) Review

How much does Cartesia cost?

The Free tier costs $0 per month and includes 20,000 credits. It also lists a $1 prepaid agent allowance. Pro costs $5 per month with 100,000 credits and a $5 agent allowance. Startup costs $49 per month with 1.25 million credits and a $49 agent allowance.

Scale costs $299 per month with 8 million credits and a $299 agent allowance. Enterprise is a custom quote with volume pricing. The public page also mentions custom concurrency limits and compliance paperwork at that level. All figures were read on 1 November 2026.

The commercial use licence starts at Pro. Free users can test the models but should not ship commercial audio under that tier. Instant voice cloning also starts at Pro, while Pro voice cloning waits for Startup. Scale adds priority support and higher concurrency, which matters for busy call flows.

What do Sonic, Ink and Managed Agents actually do?

Sonic is the text to speech engine. It takes a script and returns audio in a chosen voice. Cartesia says the model runs with low latency, which is the main requirement for a live agent. The voice library includes a range of speakers and languages.

Ink is the speech to text side. It transcribes audio as it arrives, rather than waiting for a finished file. A call agent needs that behaviour to reply before the caller loses patience. The transcript can also feed summaries and compliance logs after the call.

Managed Agents combines both models into a hosted phone agent. Cartesia prices it separately at $0.03 per minute. The platform also lists 44 languages on its language page. A developer can start with the hosted agent and later move logic to their own stack.

Who is Cartesia for?

The platform suits teams building voice agents that talk to customers in real time. Support lines, appointment booking and lead qualification are the use cases Cartesia names. Those flows need fast speech recognition and fast speech output in the same loop.

Developers who already own their orchestration can use Sonic and Ink as APIs. That route keeps prompts, tools and business logic inside the existing stack. The trade-off is that the team must handle telephony, retries and monitoring itself.

Teams that only need prerecorded narration should compare simpler text to speech services first. Cartesia's strength is conversation, not batch audio production. A podcast workflow rarely needs streaming transcription or agent minutes.

How a Cartesia voice agent comes together

Start with one narrow call flow, such as rescheduling an appointment. Write the questions the agent must ask and the answers it can accept. Keep the script short, because every extra turn adds latency and cost.

Choose a voice from the library and test it on the actual phone line. A voice that sounds good in a browser can behave differently under compression. Check how the agent handles interruptions, silence and background noise before launch.

Wire Ink for transcription and Sonic for replies, or let Managed Agents run the loop. Log each call with its duration and agent minutes. Review the first fifty calls, fix the questions that trigger confusion, then widen the scope.

Track credit burn against the monthly allowance. A busy week can consume a plan quickly, and the prepaid agent balance is separate from speech credits. Set alerts before the balance runs low, so a live phone line never goes silent.

What to check before committing to Cartesia

Confirm the licence that covers your use. Commercial audio needs Pro or higher, and the terms may differ for cloned voices. Read the acceptable use policy before you upload a customer recording or a synthetic voice sample.

Estimate both meters, not just credits. Speech generation draws on credits, while agent minutes draw on the prepaid agent balance. A quiet month on calls still consumes credits if you generate audio elsewhere.

Ask about data retention for call audio and transcripts. Voice data can include personal information, and retention rules vary by industry. Cartesia lists compliance paperwork at Enterprise, so regulated buyers should start that conversation early.

Treat vendor rankings as marketing until you test them. Cartesia promotes leaderboard positions, but your latency depends on network path, telephony provider and script design. Run a small pilot with real callers before signing an annual deal.