Cartesia

Cartesia wasn't started by a team that wrapped someone else's model in a nicer dashboard. According to its own site, the people behind it pioneered state space models (SSMs), Mamba and H-Nets, architectures pitched as an alternative to transformers for low latency, long-context work. The site doesn't say much about who the founders are individually, but it's clear about the thesis: build "AI that learns from and interacts with the world like humans do," and prove it first in real-time voice, where speed matters most.

That research background shapes the products. Sonic is the text-to-speech model, Ink handles streaming speech-to-text, and Managed Agents is the platform for building and shipping voice agents on top of both. Cartesia says it's ranked #1 on the Artificial Analysis Speech Arena and Speech to Text leaderboards, and it sells everything through "one API" meant to be fast enough for live conversation. The pitch to you is simple: if your product talks to people in real time, latency is the bottleneck, and these models were designed around that constraint from the architecture up.

The go-to-market leans toward regulated industries. Finance, healthcare and government get top billing, with examples like real-time outbound fraud verification calls, and the same models can run in the cloud via regional endpoints, on-premise or on-device, with in-region inference for data residency. Pricing starts free with 20K credits a month, and Pro is $5/mo. The catch is that commercial use and voice cloning sit behind paid tiers, and the homepage is heavier on positioning than on hard specs like latency figures or supported languages. You'll be reading the docs before you make a call.