From the source
Gradium released a new Text-to-Speech model with 50ms first-chunk latency, which it claims is the lowest among frontier TTS models.
The model beats Eleven v4 Turbo on latency while improving naturalness and expressivity across all languages, and retains handling of hard cases like phone numbers and alphanumeric text.
It is now the default model for Gradium's API and studio, out of beta.
