Cartesia Ships Sonic-3.6: A Streaming TTS Model That Now Leads Both Artificial Analysis Speech Arenas

Details

Cartesia released Sonic-3.6, the latest version of its real-time text-to-speech model, three months after Sonic-3.5. The new model leads both Artificial Analysis speech arenas: 1,283 Elo on the Provider Voice board and 1,123 Elo on the Controlled Voice board. The Controlled Voice result is more meaningful — it clones all models onto the same eight reference voices, isolating the synthesis engine from the voice catalog. Sonic-3.6 leads that board, with Sonic-3.5 second and ElevenLabs Eleven v3 third. The model is built on state space models rather than transformers — an architectural choice Cartesia frames as enabling better latency-naturalness tradeoffs. Cartesia states sub-90ms time-to-first-audio (vendor-measured model latency, not end-to-end round trips). Available in beta as a hosted API. Pricing: Artificial Analysis normalizes Sonic-3.6 at $49/1M characters — half of ElevenLabs Eleven v3 at $100, and well above budget options like Speechify Simba 3.2 at $10. Production-relevant features include inline expression tags like [laughter], instant voice cloning from ~10 seconds of audio, custom pronunciation dictionaries with IPA overrides, and native handling of alphanumerics like order numbers and phone numbers. There are no open weights and no Hugging Face repo — Sonic is a closed commercial model available through Cartesia's hosted API. Tiers range from free/solo to enterprise with DPAs, BAAs, and SSO support.