Underdog shows off a 32MB speech model in new demo

Despite its small size, Underdog's new model can generate both speech and music.

TechCrunch · 2026-11-06T13:02:16.000Z