Underdog releases Woof, a 31M-parameter open text-to-speech model that runs in a browser

Underdog has released Woof, a text-to-speech model with 31 million parameters and a size of 31.7 MB, and the model is open for anyone to download. The company says it runs on most devices without an internet connection. To show this, Underdog published a browser-based game on November 5, 2026, in which every voice is generated by Woof. TechCrunch reported the demo on November 6.

What was released

• Model: Woof, which Underdog calls its "new frontier text-to-speech model"
• Parameters: 31 million
• Size: 31.7 MB (TechCrunch's headline rounds this to 32MB)
• Availability: open for download
• Runtime: in the browser, and on most devices offline, according to Underdog
• Pricing, context window and benchmark scores: not yet disclosed

The "frontier" label is Underdog's own description. No benchmark results were published with the announcement.

The demo

The game is set in a room with a retro tech theme, including a typewriter, a CRT monitor and a desk lamp that clicks when its switch is flipped. The intro warns players about "intrusive" cloud providers in the town, which TechCrunch describes as ostensibly a jab at cloud-based models.

Players can type on the typewriter with their keyboard, and the game speaks the text back. A music tuner can be rotated to generate different kinds of music. All of this runs in the browser through Underdog's speech generation model, per the company. Underdog's post on X says players can defeat a character called Darkcloud to receive an invite.

Underdog's other releases

The Woof demo comes weeks after Underdog launched Bark, a 167 MB speech-to-text model also designed for edge devices, aimed at transcription and dictation. Underdog has also released a 27-billion-parameter model that can run on laptops. It has a waitlisted app that it positions as a locally-hosted alternative to Muse and Instinct.

Founder Sigil Wen told TechCrunch he started building after deciding that, instead of using notetakers like Granola, he wanted a system built on AI models that run on users' own devices so they own their data. "I think we will be able to run most capable models of today on the devices you already own within months. We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI," Wen said.

What this means

Woof pairs a downloadable model with a demo that runs in a browser tab. Together with Bark for speech-to-text, Underdog now offers open on-device models for both directions of voice. The company's pitch is that audio generation and transcription can run locally, so user data stays on the device.

TechCrunch notes that a model this small likely will not replace larger ones, but it may be enough to drive local assistant experiences or personalized podcasts. The claims that Woof is "frontier" and runs on most devices offline come from Underdog and have not been backed by published benchmarks. Voice quality and latency across hardware are the details to check once developers start downloading the weights.