Underdog shows off a 32MB speech model in new demo November 6, 2026 Despite its small size, Underdog’s new model can generate both speech and music.
