Cartesia
Real-time voice AI on state-space models — Sonic streaming TTS, Ink STT with native turn detection, and Line, a code-first voice-agent platform.
Cartesia builds voice AI on state-space models: Sonic streaming TTS — vendor-claimed sub-100ms model latency, 42 languages, emotion controls — Ink streaming STT with turn detection native to the model, and Line, a code-first platform for deploying voice agents with hosted infra, telephony, and evals. Freemium credits; commercial use starts at the low-cost Pro tier.
Cartesia is the latency specialist of voice AI — founded by the creators of the state-space model architecture, and betting that conversation-grade voice is a realtime systems problem. Its stack covers both directions (Sonic out, Ink in) and, with Line, the agent platform that runs them.
Highlights
- Sonic TTS — streaming-first synthesis with vendor-claimed sub-100ms model latency; Sonic 3.5 (May 2026) spans 42 languages with emotion and laughter controls.
- Ink STT — streaming transcription with turn detection native to the model (turn-start/turn-end events, no external VAD), plus careful handling of phone numbers, emails, and IDs; Ink-2 launched May 2026 (English-first).
- Line — the voice-agent platform: SDK/CLI with one-command deploys, hosted infra, provisioned phone numbers, recordings/transcripts, latency dashboards, and built-in evals.
- Voice cloning — instant (Pro) and professional tiers.
- SSM pedigree — the architecture bet (efficient streaming inference) is the product's whole thesis.
In an AI-assisted workflow
Sign up, take an API key, and stream over WebSocket — or let Line own the loop. In a voice-agent pipeline, Cartesia typically slots in as the TTS (and now STT) where time-to-first-audio defines how human the agent feels; native turn detection removes one of the pipeline's trickiest components.
NOTE
Plan mechanics worth knowing: the free tier is non-commercial (commercial use starts at Pro), credits meter TTS ~6× faster than STT, and the older T2A API was deprecated in March 2026 — build against the current endpoints.
Good to know
$64M Series A led by Kleiner Perkins (March 2025); a larger late-2025 raise is third-party-reported but not vendor-confirmed, so we don't state it. Hosted/proprietary (the GitHub org carries SDKs). Voice agents bill separately from credits: as of September 2026 the pricing page lists $0.06 per minute of call duration, plus $0.014 per minute when the call runs on a Cartesia-provisioned phone number. Against the field — ElevenLabs' breadth, Deepgram's enterprise STT, Vapi as the assemble-don't-build alternative to Line — see Best TTS APIs and Best STT APIs.
Frequently asked questions
- What makes Cartesia different from ElevenLabs?
- Architecture and posture. Cartesia's models are state-space (SSM/Mamba lineage — its founders created the architecture), built for streaming-first, low-latency realtime voice; ElevenLabs leads on voice variety, expressiveness, and its broader audio product surface. For interactive agents where every millisecond of latency is conversational quality, Cartesia is the specialist; for produced audio and voice breadth, ElevenLabs.
- What is Cartesia Line?
- Their voice-agent platform (GA August 2025): code-first SDK and CLI with one-command deploys, hosted infrastructure, telephony via Cartesia-provisioned numbers, imported Twilio numbers, or your own SIP trunk (all documented as generally available, with no beta label as of September 2026), call recording and transcripts, latency metrics, and LLM-as-judge evals — defaulting to Sonic + Ink, so the whole loop runs on Cartesia's stack at per-minute pricing.
- Is there a free tier?
- Yes — plans as of September 2026 from cartesia.ai/pricing: Free is $0/mo with 20K credits a month, but it is non-commercial. A commercial-use license starts at Pro ($5/mo, 100K credits), which also unlocks instant voice cloning; Startup is $49/mo for 1.25M credits, Scale is $299/mo for 8M credits, and Enterprise is custom. Credits meter both TTS and STT (TTS burns them several times faster).
Filed under
tts · stt · voice · realtime · voice-agents
Related
- Best Speech-to-Text APIs in 2026The STT field, honestly ranked — Deepgram and AssemblyAI's hosted duel, Whisper as the open baseline, Cartesia Ink for latency — and how to pick by workload.
- Best Text-to-Speech APIs in 2026The TTS APIs worth building on — ElevenLabs for quality and breadth, Cartesia Sonic for realtime latency — and how to choose for agents vs produced audio.
- Realtime Voice Agents: Build on LiveKit, Buy Vapi, or Pipeline with PipecatThe three ways to ship a realtime voice agent in 2026 — open infrastructure, managed platform, or OSS pipeline — and how speech-to-speech models fit in.
- DeepgramA voice-AI platform with fast, accurate speech-to-text (Nova) and low-latency text-to-speech (Aura), plus a bundled Voice Agent API.
- ElevenLabsA voice-AI platform for high-quality text-to-speech, voice cloning, dubbing, and real-time conversational agents, via API.
- AssemblyaiSpeech AI platform: promptable Universal-3 Pro STT, a flat-rate Voice Agent API, and speech understanding — summarization, sentiment, PII redaction.
- How to Build a Voice Agent: The STT → LLM → TTS PipelineHow to build a real-time voice agent: the STT → LLM → TTS pipeline, the latency budget that makes or breaks it, and how to wire each stage.
- LivekitOpen-source realtime infrastructure — a WebRTC server plus the LiveKit Agents framework for production voice AI, with turn detection, telephony, and cloud.