Cartesia
NewCartesia offers Sonic text-to-speech and Ink speech-to-text models through a low-latency API, with instant and pro voice cloning, 44-language speech output and Managed Agents for building phone voice agents.
Cartesia
Explore features, practical uses and pricing below.
Cartesia is a voice AI company that sells its own speech models to developers. Its main product is Sonic, a streaming text-to-speech model. It also offers Ink for speech to text and Managed Agents, a layer for building phone and app voice agents on top of both. The company says its models run on state space models, an architecture it ties to low latency at scale.
Sonic and Ink, the two models underneath
Sonic 3.6 is the current text-to-speech model. The TTS model docs say it speaks 44 languages, from English, Hindi and Japanese to Tamil, Odia and Urdu, and that it adapts pacing and intonation without SSML tags. The docs home describes streaming at sub-90 ms latency. Calling the model ID sonic-3.6 tracks the latest stable snapshot, a dated ID such as sonic-3.6-2026-08-27 pins one release, and sonic-preview is for testing upcoming changes. Pinning matters if a product's voice must sound the same next quarter.
Ink 2 handles speech to text. Cartesia describes it as a streaming model with native turn detection, meaning it marks when a speaker starts and stops. In a voice agent that decides when the bot should reply, which is often what separates a natural call from an awkward one.
Voice cloning from ten seconds or thirty minutes
Instant cloning needs 10 seconds of audio to start, and the cloning guide says up to 60 seconds helps keep the speaker's accent on Sonic 3.6 and newer. Each clone is built from one language, recorded in a language the speaker actually speaks, and further languages are added afterwards through an accents API. Pro voice cloning trains on 30 minutes or more of one speaker and is meant for rare accents or character voices that an instant clone might flatten. Clips can be recorded or uploaded in the dashboard, or sent to the Clone Voice API.
Building a phone agent with Managed Agents
Managed Agents combines Ink, an LLM you choose and Sonic, while Cartesia handles listening, turn-taking and interruptions. The agents docs describe a flow like this:
- Create the agent in the playground or API, pick the LLM and write its instructions.
- Add tools: system tools such as ending a call, webhook tools that call your server, or client tools your app runs.
- Connect a phone number and publish, since live calls use the published version.
- Review call quality with built-in and custom metrics, then revise the prompt.
Teams with their own orchestration can skip this layer and call Sonic and Ink directly.
Who Cartesia is built for
Cartesia is aimed at developers and enterprises building real-time voice products: support lines, in-app assistants and voice features on devices. The homepage lists regional cloud endpoints plus on-premise and on-device deployment, which can matter for data residency. Creators who want a voiceover editor with dubbing and sound effects will find more studio tooling in ElevenLabs. For transcription-heavy work, Deepgram is a natural comparison. Other speech output tools are collected in our text to speech category.
Credits, plans and concurrency
The pricing page lists a Free plan with 20K credits a month, Pro at $5 a month with 100K credits and instant cloning, Startup at $49 with 1.25M credits and two pro voice clone slots, Scale at $299 with 8M credits and four slots, and custom Enterprise terms. Each plan also caps concurrency, from 2 simultaneous TTS requests on Free to 15 on Scale. Agent calls are billed per minute on top of that, with LLM usage billed separately. Commercial use is listed from Pro upward.
Cartesia questions developers often raise
Does Cartesia have a free plan?
Yes. The Free plan includes 20K credits a month and two concurrent TTS requests. Instant voice cloning and commercial use are listed from the Pro plan.
Can a cloned voice speak other languages?
Yes. The docs say to clone in the speaker's own language first, then add languages and accents to that clone.
Do I have to use Managed Agents?
No. Sonic and Ink work on their own for narration, captions or any app that needs speech in or out.