Cartesia · 2024–26 · SSM
Sonic uses a state-space model rather than a transformer, so inference cost scales linearly with length. Sonic 3.5 (May 2026) is the latency specialist: vendor figures around 40–82 ms to first audio. The clip here is a published marketing demo of laughter / naturalness from cartesia.ai/sonic, not our shared script.
The lowest published time-to-first-audio among flagship APIs.
Vendor / paper demo
Sonic (laughter / non-speech demo) · Official Cartesia Sonic product-page naturalness/laughter demo.
Vendor demo: Cartesia Sonic product-page clip of generated laughter / non-speech. This is not a spoken script; listen to the file.