The model is available on Hugging Face under the Fish Audio Community License and via an API priced at $15 per 1 million characters, compared to $80 per 1 million for Eleven v4. Designed for long-form content, the system enables users to adjust the emotion and tone of individual words using natural language instructions. Fish Audio currently serves more than 15M users and recently completed a $52M seed funding round.
Tuesday, October 20, 2026
Story
Fish Audio Drama 3 Leads All TTS Models in Pronunciation Robustness
AI7 days ago
The open weight system from Fish Audio has secured first place across three text to speech rankings from Artificial Analysis. Its Drama 3 Preview model scored 93.2% in Pronunciation Robustness, surpassing the scores of the closed source Eleven v4 and Eleven v4 Turbo models, which recorded 91.7% and 90.1% respectively. The 35B parameter mixture-of-experts model supports 83 languages and reached a Provider Voice Elo of 1,243, a 101 point gain over the company's previous S2.1 Pro.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.