Having built with and tried every voice model over the last three years, real time and non-real time... this is off the charts compared to anything I've seen before.

And open weight too! So grateful for this.