# Underdog Releases 163 MB Speech Model

> Underdog Bark 3.0, a 163 MB speech model, matches larger systems on recorded-meeting transcription.
- Title: Underdog Releases 163 MB Speech Model · RuntimeWire | Shortify
- Summary: Underdog Bark 3.0, a 163 MB speech model, matches larger systems on recorded-meeting transcription.
- Keywords: underdog, bark, conway research, speech model, nvidia, parakeet, cc-by-4.0, technology, Releases, 163, Speech, Model
- Source: RuntimeWire — https://runtimewire.com/article/underdog-bark-3-speech-model
- Author: RuntimeWire
- Published: 2026-10-21T05:16:08.953+00:00
- Read time: 1 min
- Topics: underdog, bark, conway research, speech model, nvidia, parakeet, cc-by-4.0, technology, ai
## Underdog Bark 3.0
Sigil Wen released Underdog Bark 3.0 on October 21st, a 163 MB speech-to-text model for laptops and phones.
## Competitive Accuracy
The model matches or beats larger systems on recorded-meeting transcription while running locally.
## Permissive License
Model weights are available under the CC-BY 4.0 license for on-device use.
## Key takeaway

Underdog Bark 3.0, a 163 MB speech model, matches larger systems on recorded-meeting transcription.