Underdog releases 32MB on-device speech model Woof
What happened
On 6 November 2026, TechCrunch reported that Underdog has released a browser game demo built on its new speech model Woof. The model is just 32MB with 31 million parameters, and the company says it can run without a network on most devices; it is also available for download. In the demo, users can type any text on a typewriter and have the model read it aloud, or turn a tuner to generate different music. The report notes that weeks earlier Underdog also launched Bark, a 167MB on-device speech-to-text model. Founder Sigil Wen said he hopes to run most of today's more capable models on users' existing devices within months. So far the demo is live and the model is available for download, with no further reporting on deployment.
Written by AI from the reports · updated 1 h ago
Timeline
Follow the reports to see the event from each side.
- TechCrunch · AIUnderdog shows off a 32MB speech model in new demo
Despite its small size, Underdog's new model can generate both speech and music.
Heat of this event
Not enough continuous observations yet to draw a trend.