Friday, November 6, 2026
Tech Beat
Nov 6, 2026, 1:02 PMArtificial Intelligence

Underdog Debuts 31.7MB Woof Voice Model for Offline AI

Underdog debuts Woof, a 31.7MB, 31-million-parameter voice model that runs offline in browsers, joining its 167MB Bark model for private AI on devices.

Listen to this briefingAudio briefing

Summary

Underdog introduced Woof on November 5, 2026, a 31.7MB, roughly 32MB, text-to-speech model with 31 million parameters. The downloadable open model runs inside a browser and, Underdog says, works offline on most devices. Its retro game lets users type text for Woof to speak, rotate a tuner to generate music and click a sounding desk lamp. All voices use Woof, while warnings about intrusive cloud providers and a Darkcloud challenge for an invite promote local processing. The compact model is unlikely to replace larger systems but could power local assistants and personalized podcasts.

Woof follows Bark, Underdog’s 167MB speech-to-text model for edge transcription and dictation, launched weeks earlier. Founder Sigil Wen began developing his system instead of using notetakers such as Granola, seeking on-device AI that gives users ownership of their data. He expects most currently capable models to run on existing consumer devices within months and wants Underdog to produce capable, open, American-made models for private AI. The startup has also released a 27-billion-parameter laptop model and opened a waitlist for a locally hosted app positioned against Muse and Instinct.

Positives

  • Woof occupies 31.7MB and uses 31 million parameters, making browser-based text-to-speech practical on constrained devices.
  • Underdog has opened Woof for anyone to download and says it operates without an internet connection on most devices.
  • The browser demo generates speech, music and sound locally through interactive typewriter, tuner and lamp controls.
  • Bark adds a 167MB edge model for private transcription and dictation alongside Woof’s speech generation.
  • Underdog’s 27-billion-parameter model extends its local AI strategy to more capable laptop hardware.

Risks & concerns

  • Woof’s small size means it is unlikely to replace the larger speech models competing on capability.
  • Offline operation across most devices is Underdog’s claim, and no comparative performance benchmarks were provided.
  • Access to Underdog’s locally hosted alternative to Muse and Instinct remains restricted by a waitlist.
  • The game directs users to defeat Darkcloud for an invite, adding an access hurdle despite Woof itself being downloadable.
Primary sourceTechCrunchhttps://techcrunch.com/2026/11/06/underdog-shows-off-a-32mb-speech-model-in-new-demo/
Read full article
Editorial note: Tech Beat summarizes and analyzes third-party reporting. The source link is the authoritative article. This page does not reproduce the full source text.

More From The Wire

Artificial IntelligenceNov 6

Open ASR Leaderboard Adds Egyptian and Nigerian English Tests

Artificial IntelligenceNov 5

Airbnb AI Travel Agent Set for First Half of 2027 Launch

Artificial IntelligenceNov 5

Google Gives Schools Gemini Controls but Refuses AI-Free Search