Changelog

simba-3.0 is now available on POST /v1/audio/speech and POST /v1/audio/stream via the model field. It’s the new streaming-native voice model with lower TTFB and richer expressivity, including direct support for per-voice speaking-rate and ADV (Arousal, Dominance, Valence) emotion controls inherited from the voice catalog.

1
{
2
  "input": "Hello, world!",
3
  "voice_id": "george",
4
  "model": "simba-3.0"
5
}

Currently English only — multilingual coming soon

simba-3.0 currently supports English voices only. Requests with a non-English voice return 400 with the rejected locale called out in the message. Multilingual support is coming soon; the model name stays simba-3.0 across that change, so no migration is required.

For non-English voices today, continue to use simba-multilingual.