Cartesia
Ultra-low-latency voice AI on Mamba SSMs
Overview
Cartesia builds voice-AI models on state-space architectures (Mamba) for ultra-low latency — sub-90ms time-to-first-byte. Used in voice agents and live translation. Founded by Mamba's original authors.
Use cases
What you can produce with Cartesia
- Add real-time text-to-speech to a voice agent with roughly 90ms model latency using the Sonic API and streaming websockets.
- Clone a voice from a short audio sample and use it for consistent branded speech across an application.
- Generate expressive speech with controllable emotion, pacing and even laughter in 42 languages with Sonic-3.
- Transcribe live audio with the Ink speech-to-text models to build a full speech-in, speech-out pipeline on one platform.
- Deploy voice models on-premise or on-device for latency-critical or data-sensitive workloads.
- Integrate Cartesia voices into agent frameworks and telephony platforms such as Vapi for phone-based AI agents.
- Meet enterprise compliance requirements with SOC 2 and HIPAA-compliant voice infrastructure.
Cartesia's Sonic models are a leading choice for real-time text-to-speech, prized for very low latency and natural prosody, making them well-suited to voice agents and interactive applications. The API is developer-friendly with a usable free tier (20K credits), pay-as-you-go at roughly $50 per million characters, voice cloning, and clear pricing tiers from Free to Enterprise.
It's an infrastructure product, so non-developers won't use it directly, and heavy voice-agent usage adds per-minute costs on top of TTS credits. Multi-language support is growing but English remains strongest; it's globally accessible as an API with no ASEAN-specific provisions. An excellent pick for engineers building latency-sensitive voice experiences.
What people say
Cartesia is one of the fastest-rising voice-AI infrastructure companies of the 2025-2026 cycle. Founded by the researchers behind the Mamba state-space architecture, it raised a $64M Series A in March 2025 and a $100M Series B in November 2025 (backers include Kleiner Perkins, Index, Lightspeed and NVIDIA), bringing disclosed funding to roughly $191M. The product line has moved quickly: Sonic-3, launched with the Series B, supports 42 languages with around 90ms model latency and added laughter and emotional expression; Sonic-3.5 and the Ink-2 speech-to-text model followed in June 2026 as a unified real-time voice stack. The company reports over 10,000 customers, including Quora, Cresta and Rasa.
Developer sentiment centers on speed. The sub-100ms time-to-first-audio is repeatedly called out as the differentiator for real-time voice agents, where competitors' latency makes conversations feel laggy. Reliability (99.9% uptime claims, SOC 2 and HIPAA compliance, on-prem and on-device options) and strong voice cloning also earn praise, and Cartesia's own blind tests claim listeners preferred Sonic-3 over ElevenLabs roughly 62% to 39% - a vendor-run number, but consistent with the generally positive developer chatter around the launch.
The caveats: this is a developer API, not a consumer app, so there is little of the G2/Capterra review footprint that end-user tools accumulate, and independent long-term comparisons are still thin. Developers in community discussions note occasional pronunciation quirks and that ElevenLabs still has a broader voice library and ecosystem, and costs at high volume require careful modeling.
Cartesia genuinely fits engineering teams building voice agents, live translation, IVR replacements or any latency-sensitive speech product. Individual creators who just want a simple narration tool with a polished web UI are better served by consumer-oriented TTS platforms.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Cartesia's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Cartesia unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Cartesia directly →
Spotted something out of date? Suggest an update →
More in Video & Audio
