FULL STORY

Cartesia Sonic-3.6: From Launch to Top of the Voice Arena

Cartesia launched Sonic-3.6, a real-time TTS model supporting 44 languages. It soon topped the Voice Arena leaderboard, beating ElevenLabs in head-to-head evaluations.

2026-08-18 ~ 2026-08-28 · 2 episodes · 9 posts

Episode 1 · Cartesia Releases Sonic-3.6, Topping Speech Leaderboards (2026-08-18, 4 posts)

Cartesia has launched Sonic-3.6, a TTS model covering 44 languages with improved naturalness, topping both Artificial Analysis speech leaderboards with an Elo of 1123 and generating nearly 3x faster than Eleven v3.

Episode 2 · Cartesia's Sonic-3.6 TTS Launches, Topping Voice and Audio Leaderboards (2026-08-28, 5 posts)

On August 28, Cartesia released its real-time TTS model Sonic-3.6 and announced general availability, immediately topping several speech synthesis leaderboards: it ranked first on the Voice Arena American English leaderboard with 1085 Elo, tied with Google DeepMind's Gemini 3.1 Flash TTS, and also took first place on both the controlled and provider speech tracks of the Artificial Analysis TTS leaderboard. The team said it received positive feedback since the preview.

Confirmed

  • Sonic-3.6 became generally available on 08-28 and ranks first on the Voice Arena American English leaderboard (1085 Elo, tied with Gemini 3.1 Flash TTS)
  • It ranks first on both the controlled and provider speech tracks of the Artificial Analysis TTS leaderboard
  • The Cartesia team said it decided in January this year to stop tuning the existing paradigm, rebuilt the architecture from scratch, and surpassed its predecessor Sonic-3.5 within two months

Independent Testing

Blogger @socialwithaayan compared Cartesia with ElevenLabs in real-world scenarios: both were asked to read out the same flight notification containing detailed information such as flight numbers, booking references (PNR), times, and gates (e.g., EK 418 / W8R4U). The results showed Cartesia clearly outperformed in clarity and naturalness when reading such detailed information, with a significant gap.

Why It Matters

The TTS space has long been dominated by vendors like ElevenLabs. Cartesia reached the top through an architectural rebuild rather than fine-tuning an old paradigm, and it performs notably well in real-world announcement scenarios involving numbers and codes that are prone to errors—directly relevant for practical uses such as customer service notifications and travel announcements.