Cartesia Debuts Sonic‑3.6, a 44‑Language Voice Model That Tops Third‑Party Leaderboards

Voice AI startup Cartesia has launched a beta of its new text‑to‑speech model Sonic‑3.6, which supports 44 languages and has jumped to the top of independent voice quality leaderboards run by Artificial Analysis, according to the latest AI news rundown. Compared with traditional TTS systems, Sonic‑3.6 is designed to produce more natural intonation, breathing, and pacing, reducing the “robotic” feel even over longer recordings. Beyond broad language coverage, Cartesia is emphasizing fine‑grained style control: developers can adjust speed, energy, and expressiveness so the same script can be delivered as a calm support message, an upbeat marketing spot, or a formal documentary narration. The beta is currently available via an API for developers, with some early partners allowed to test commercial workloads such as call center voice bots, video voice‑overs, and e‑learning content. As with other advanced voice models, the launch comes amid heightened concern over abuse of voice cloning for fraud and impersonation. Cartesia says it is restricting certain use cases and tying high‑fidelity voices to verified talent, but how robust its safeguards prove in practice will be crucial in determining whether high‑end generative voice becomes a trusted building block for global AI applications.