Cartesia

Cartesia is a real-time voice AI platform built around Sonic, its streaming text-to-speech model. It is made for developers and teams building voice agents, live assistants, and interactive apps that need natural speech with very low latency. You reach the models through an API and SDKs, and can run them in the cloud, on-premise, or on-device.

Key Highlights

• Sonic-3.5 streaming text-to-speech with expressive voices in 40+ languages
• Ink-2 speech-to-text for transcription in voice pipelines
• Line voice agents that handle live phone and in-app conversations
• Instant voice cloning from a short audio sample
• Deploy in the cloud, in your own VPC or hardware, or on-device
• SDKs and developer tools for production integration

What Makes It Different

Cartesia's models are built on State Space Models (SSMs), an architecture its founding team helped pioneer at Stanford (including Mamba and H-Nets). SSMs are designed for live, synchronous interactions, so Sonic targets ultra-low time-to-first-audio rather than batch generation. Sonic-3.5 streams its first audio in roughly 90 milliseconds, fast enough for back-and-forth conversation where any delay is noticeable.

The other differentiator is deployment flexibility. The same models and agents run across cloud, on-premise, and on-device, with inference kept in-region for teams with data residency, compliance, or latency needs a single cloud endpoint cannot meet.

Features & Capabilities

The core workflow is API-first: send text to Sonic and stream audio back, send audio to Ink-2 for a transcript, or combine both with the Line agent layer for full voice conversations. Agents can take phone calls on a Cartesia-provided number and connect to your own systems and logic at scale.

Beyond synthesis, it offers instant voice cloning from a short sample, professional voice cloning on higher tiers, a voice changer, and voice localization across languages. Every plan includes unlimited seats and voice slots, with concurrency and agent limits that scale by tier.

User Ratings and Testimonials

Cartesia is best known for speed. Reviewers consistently rank Sonic among the lowest-latency text-to-speech options for real-time agents, and its instant voice cloning from a few seconds of audio draws frequent praise. The common criticism is that for long-form, expressive narration, rivals such as ElevenLabs often rate higher on voice realism, so Cartesia suits live, conversational use more than polished voiceover.

Pricing & Value

• Free: $0/month, 20K credits and $1 of prepaid agents, with text-to-speech and speech-to-text
• Pro: $4/month, 100K credits and $5 prepaid agents, adds a commercial-use license and instant voice cloning
• Startup: $39/month, 1.25M credits and $49 prepaid agents, adds professional voice cloning and organizations
• Scale: $239/month, 8M credits and $299 prepaid agents, adds priority support and high concurrency
• Enterprise: custom pricing with volume rates, custom concurrency, SSO, and compliance agreements

Voice agent calls are billed at $0.06 per minute, plus $0.014 per minute for telephony on a Cartesia number. Yearly billing saves 20%, and the free tier is enough to prototype before you commit.