# Cartesia

> Cartesia builds real-time voice AI on state space models: Sonic-3.6 text-to-speech, Ink-2 speech-to-text and the Line agent platform, as of October 2026.

Source: https://metavert.io/cartesia  
Published: 2026-10-21  
Updated: 2026-10-21

**Cartesia** is a voice-AI company whose models are built for real-time conversation, where the delay before the first sound matters more than anything else. It was founded in 2023 by Karan Goel (CEO), Albert Gu, Arjun Desai and Brandon Yang, who met as PhD students in Christopher Ré's group at the Stanford AI Lab, with Ré as a co-founder and advisor. That group's research on [state space models](https://metavert.io/state-space-models) (SSMs), an alternative to the [transformer](https://metavert.io/transformer-architecture) with lower latency and longer memory, is the technical bet behind the company. Cartesia raised a $27 million seed round led by Index Ventures in December 2024 and a $64 million Series A led by Kleiner Perkins in March 2025.

### Products (October 2026)

| Product | What it does | Latest release |
| --- | --- | --- |
| Sonic | Streaming [text-to-speech](https://metavert.io/text-to-speech) in 44 languages and 61 locales, with about 60 emotion values, a \[laughter\] tag and speed and volume controls | Sonic-3.6, GA August 27, 2026 (the line began in May 2024) |
| Ink | Streaming speech-to-text for agents, with turn events and "semantic endpointing"; five languages | Ink-2, July 9, 2026 |
| Line | A code-first platform for building voice agents: any LLM, SIP (beta), Twilio, batch calling and evaluations | Launched August 19, 2025 |
| Voice cloning and Voice Changer | Instant clones from about 10 seconds of audio; professional clones on higher plans; voice-to-voice conversion | Professional cloning May 2025 |

Cartesia markets Sonic-3.6 on first audio in under 90 milliseconds. It says listeners preferred it in up to 93% of blind head-to-head tests across fifteen locales, and it claimed first place on Artificial Analysis's speech arena at launch. These are vendor claims, and [ElevenLabs](https://metavert.io/elevenlabs) claimed the same arena's top spot a month later for Eleven v4. Rankings move with every release.

### Pricing and Deployment

Cartesia sells monthly credit plans. Free (20,000 credits, no commercial use) is followed by Pro at $5, Startup at $49 and Scale at $299 a month, with speech at one credit per character. At list prices its plans include more characters per dollar than ElevenLabs': the $299 Scale plan carries 8 million credits. Its deployment options are broad. Sonic can run in a customer's own cloud through Baseten, Amazon SageMaker or Together AI, self-hosted Sonic-3.6 became available to enterprise customers in September 2026, and Line offers zero data retention on enterprise plans. That suits call centers and regulated industries that can't send audio to a shared API.

### How It Compares

Cartesia is a specialist. ElevenLabs is far broader, covering narration, transcription, dubbing, music and cleanup under one account, and it's the usual choice for produced audio and [AI dubbing](https://metavert.io/ai-dubbing). Cartesia is the usual choice where speech has to feel immediate: phone agents, assistants and other [conversational AI](https://metavert.io/conversational-ai). [Deepgram](https://metavert.io/deepgram) now competes on both ends of a voice agent with Flux recognition and Flux TTS. Open-weight text-to-speech models compete for teams that want to host their own. Easy cloning from a short sample brings the same [voice-deepfake](https://metavert.io/deepfakes) risk at every vendor, so a consent record for each cloned voice is good practice. For the head-to-head, see [ElevenLabs vs Cartesia](https://metavert.io/compare/elevenlabs-vs-cartesia).

## Related Topics

- [ElevenLabs vs Cartesia](https://metavert.io/compare/elevenlabs-vs-cartesia) — Breadth versus real-time latency
- [Text-to-Speech and AI Voiceover](https://metavert.io/text-to-speech) — How neural TTS works and who sells it
- [Conversational AI](https://metavert.io/conversational-ai) — Where latency decides the experience
- [State Space Models](https://metavert.io/state-space-models) — The architecture behind Cartesia's models
- [ElevenLabs](https://metavert.io/elevenlabs) — The broad audio platform Cartesia competes with
- [Deepgram vs AssemblyAI](https://metavert.io/compare/deepgram-vs-assemblyai) — Speech-to-text APIs for the input side
- [Speech & Voice AI](https://metavert.io/speech-voice-ai) — The broader field
- [Deepfakes](https://metavert.io/deepfakes) — Why voice cloning needs consent

## Further Reading

- [Sonic-3.6](https://cartesia.ai/blog/sonic-3.6/) — Cartesia; the August 2026 release, latency and deployment options
- [Ink-2](https://cartesia.ai/blog/ink-2) — Cartesia; the July 2026 speech-to-text model
- [Cartesia changelog (2026)](https://docs.cartesia.ai/changelog/2026) — Sonic-3.6 GA, self-hosting and Line updates
- [Cartesia pricing](https://cartesia.ai/pricing) — plans and credits
- [Cartesia blog](https://cartesia.ai/blog) — seed and Series A announcements and model launches