Cartesia is a low-latency text-to-speech and voice-AI API built for developers who need real-time voice agents, not just offline narration. Its Sonic TTS model claims sub-90ms time-to-first-audio-byte, and the platform bundles speech-to-text (Ink) and a voice-agent orchestration layer (Line) alongside TTS so a full conversational pipeline can run on one stack. Pricing is credit-based with a genuine $0/month free tier and paid plans starting at $5/month, aimed at teams building phone- or app-based voice agents rather than casual narration users.
Who builds it
Cartesia was founded by Stanford AI Lab researchers Karan Goel (CEO), Albert Gu, Arjun Desai and Brandon Yang, who invented State Space Models (SSMs) — the architecture Cartesia uses instead of Transformers to get faster, more efficient speech generation. The company has raised roughly $191M total, including a $100M round with participation from NVIDIA, Kleiner Perkins, Index Ventures and Lightspeed.
Core features
- Sonic text-to-speech model with sub-90ms claimed latency for real-time conversational use
- Natively multilingual voice generation across 40+ languages, with instant voice cloning from about 10 seconds of reference audio
- Cloned voices can be localized into 42 languages while preserving tone and speaker identity
- Ink speech-to-text model for streaming transcription, including turn-detection (deciding when a user has finished speaking)
- Line: a voice-agent platform for building and deploying phone/voice bots, billed separately at $0.06/minute for agent call time plus $0.014/minute for Cartesia-provided phone numbers
- Deployment flexibility via cloud API, on-premise or on-device models, alongside REST API access documented at docs.cartesia.ai
Pricing
- Free: $0/month — 20,000 credits (roughly 27 TTS minutes or ~1h51m of STT), 1 voice-agent slot, 8 concurrent calls; no commercial-use license
- Pro: $5/month — 100,000 credits, adds a commercial-use license and instant voice cloning, 3 agent slots, 12 concurrent calls
- Startup: $49/month — 1.25M credits, professional voice cloning and organization features, 5 agent slots, 20 concurrent calls
- Scale: $299/month — 8M credits, priority support and higher concurrency, 10 agent slots, 60 concurrent calls
- Enterprise: custom pricing with custom credit/concurrency limits, DPAs/BAAs, SSO and dedicated support
- All paid tiers additionally meter voice-agent minutes ($0.06/min) and Cartesia phone numbers ($0.014/min) on top of the monthly credit allotment; check cartesia.ai/pricing for current numbers since usage-based pricing can change
Who it's for
Developers prototyping a voice agent or TTS feature can start entirely free on the 20K-credit plan, though the Free tier lacks a commercial-use license, which the $5/month Pro plan adds along with instant voice cloning. Teams shipping production phone or in-app voice agents at real call volume are the target for Startup and Scale, where per-minute agent/telephony costs and higher concurrency limits become the deciding factor over raw credit counts.