Cartesia tops both voice leaderboards: Sonic-3.6 at 90ms TTS, Ink-2 at 100ms STT

rohanpaul_ai · x · 2026-09-08

Cartesia released Sonic-3.6 (text-to-speech, 90ms latency) and Ink-2 (speech-to-text, 100ms transcript latency), claiming the #1 streaming spot for both speaking and listening models.
The architecture shift: the voice stack is becoming part of the agent's execution loop, with listening and speaking paths combined around it.
Voice agents are systems where 100ms in the wrong place is very noticeable, so Cartesia is attacking both sides of the loop at once.

More from Models