Overview
Cartesia is a voice-AI infrastructure company founded in 2023 by ex-Google researchers, built around a custom model family it calls Sonic. The core pitch is real-time, low-latency speech: where many TTS providers add hundreds of milliseconds of delay, Cartesia targets conversational speeds suitable for live phone calls, in-app voice assistants, and on-device use. Beyond text-to-speech, the platform ships speech-to-text (its Ink model) and a hosted Voice Agents product that lets developers wire a full inbound or outbound call flow without standing up their own telephony stack. It is squarely a developer tool, you interact through APIs, SDKs, and a playground rather than a consumer app. In our evaluation Cartesia’s differentiator is latency and concurrency headroom: it is designed for systems that must hold thousands of simultaneous voice sessions, which is exactly the gap most general-purpose voice models leave open. The trade-off is that it is not a turnkey content studio, you bring the product around it.
Key Features
- Sonic real-time TTS — sub-second, streaming synthesis tuned for natural conversational cadence rather than slow batch rendering.
- Instant and professional voice cloning — clone a voice from a short sample (Pro) or with higher fidelity (Startup plus) for branded agents.
- Ink speech-to-text — a separate ASR model for transcribing calls and streaming audio in the same pipeline.
- Hosted Voice Agents — build inbound or outbound conversational agents with telephony, per-minute billing, and concurrent-call limits handled for you.
- Generous concurrency tiers — plans scale from a few parallel requests to 60 plus concurrent calls, aimed at production call centers.
- Developer-first delivery — REST and WebSocket APIs, SDKs, and a playground, with no consumer editing UI to learn.
Pricing
| Plan | Price | For |
|---|---|---|
| Free | $0/mo | 20K credits/mo, TTS plus STT, 1 agent slot, prototyping |
| Pro | $5/mo | 100K credits/mo, commercial license, instant voice cloning |
| Startup | $49/mo | 1.25M credits/mo, pro cloning, orgs, 5 agent slots |
| Scale | $299/mo | 8M credits/mo, priority support, 15 concurrent calls |
| Enterprise | Custom | Volume pricing, SSO, DPA or BAA, custom concurrency |
Comparison
Compared to ElevenLabs, Cartesia trades some of the flagship voice realism and the massive consumer ecosystem for lower latency and stronger concurrency, better for live agents, weaker as a narration studio. Against PlayHT, the two are close on developer APIs, but Cartesia bundles hosted Voice Agents and telephony that PlayHT leaves to you. Versus Hume AI, Cartesia is narrower (voice only, no emotion modeling) yet simpler to deploy for plain speech synthesis at scale.