Cartesia

Cartesia builds real time voice models and a voice agent platform, engineered from first principles on State Space Models, an architecture pioneered by its team out of the Stanford AI Lab, including Mamba and H-Nets. State space models deliver ultra low latency and efficiency at scale, which is what live, synchronous voice conversations require. The company raised a sixty four million dollar Series A led by Kleiner Perkins and remains independent. Its three products work together: Sonic, a text to speech model that streams the first byte of audio in around ninety milliseconds, with a turbo variant near forty milliseconds, across forty two languages with instant and professional voice cloning and emotional delivery; Ink, a streaming speech to text model with native turn detection that decides when a user is done speaking; and Line, a platform for building and shipping enterprise voice agents on top of those models, with built in telephony, call analytics, and observability.

Because Sonic and Ink are purpose built for conversational latency, agents respond before a caller registers a pause, which is the difference between speech that feels natural and speech that feels robotic. Voices can be cloned from ten seconds of audio and localized into dozens of languages while preserving speaker identity, and pinned model snapshots let teams lock a version in production. The same models and agents run across cloud, on premise, and on device, with inference in region to meet latency, data residency, and compliance needs.

Pricing is usage based with monthly tiers: a free plan for prototyping, Pro from about four dollars a month, Startup at thirty nine dollars a month, Scale at two hundred thirty nine dollars a month, and Enterprise through sales, with unlimited workspace seats and voice slots on every plan. Text to speech bills one credit per character, speech to text one credit per second, and voice agents by the minute, so production costs scale with call volume.