Cartesia Sonic 3
Streaming TTS on state-space models — 42 languages, natural laughter, sub-100ms streaming.
About
Sonic 3 is Cartesia's state-space-model (SSM) text-to-speech system, launched in fall 2025. It targets the hard production trio — quality, latency, and scale — with natural laughter and emotion, fine-grained control over speed/volume/emotion via API parameters and SSML tags, instant voice cloning from about 10 seconds of audio, custom pronunciation dictionaries (including IPA overrides), and native multilingual support across 42 languages with the same voice.
Cartesia claims sub-90ms model latency, but that is vendor-measured model inference, not round-trip: independent measurement by Vapi's Humanness Index (50-run round-trip test) found a 166ms median, and Cartesia's own launch disclosure put the median around 190ms — both still fast, just not sub-90ms end-to-end. The Sonic-3.6 beta (Aug 2026) sat atop the Artificial Analysis TTS quality leaderboard. Sonic 3 is available through Cartesia's API and on AWS SageMaker JumpStart; Cartesia sells credits, with Scale at $299/month for roughly 10,667 TTS minutes.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Independent: Vapi Humanness Index median 166ms TTFB (round-trip, 50 runs); Artificial Analysis TTS quality leaderboard top spot for Sonic-3.6 (Aug 2026).
Provider-reported: sub-90ms model latency, 42 languages, natural laughter.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Launch and specs verified via AWS's official SageMaker announcement plus independent latency measurement (Vapi) that honestly contextualizes the vendor's sub-90ms claim. Docked for: (1) the headline latency figure is vendor-measured model inference, ~2x slower in independent round-trip tests; (2) proprietary API-only; (3) credit pricing is harder to compare than per-character pricing.