{"id":"bark-3-underdog","slug":"bark-3-underdog","name":"Underdog Bark 3.0","orgId":"underdog-ai","family":"bark-3-underdog","releaseDate":null,"availability":"announced","openWeights":true,"license":null,"paramsTotal":null,"paramsSource":null,"contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://x.com/UnderdogAI/status/2112733007484518470","hfRepo":null,"modalities":["audio"],"aliases":["Underdog Bark 3.0","underdog bark 3.0","underdog-bark-3.0","underdog bark 3","conwayresearch/underdog-bark-3.0","underdog bark 3.0 plus","conwayresearch/underdog-bark-3.0-plus","underdog bark 3.0 pro","conwayresearch/underdog-bark-3.0-pro","underdog bark 3.0 clean","conwayresearch/underdog-bark-3.0-clean"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-21","announced":"2026-10-20","blurb":"Underdog's small open English speech-to-text model, built to run on Macs and phones. It is a 0.6B fine-tune of NVIDIA Parakeet TDT 0.6B v3, packed into a 163 MB download, and powers dictation and meeting notes in the Underdog app. CC BY 4.0. Speech-to-text, not text-to-speech, and not Suno's Bark.","api":"none","openWeights":"yes","params":{"text":"About 0.6B (163 MB download)","source":"Underdog's Hugging Face model card"},"facts":[{"label":"Task","value":"Speech-to-text, English only"},{"label":"Base model","value":"NVIDIA parakeet-tdt-0.6b-v3, via moondream/parakeet-ultra"},{"label":"Family","value":"Bark 3.0 (163 MB), Bark 3.0 Plus (180 MB), Bark 3.0 Pro (1.25 GB, full precision) and Bark 3.0 Clean (180 MB, drops fillers and false starts)"},{"label":"License","value":"CC BY 4.0 (same as the Parakeet base)"},{"label":"Weights","value":"Hugging Face org ConwayResearch; repos created Oct 15, 2026 (UTC)"},{"label":"Runtimes","value":"PyTorch loader, MLX on Apple silicon, Core ML; native in the Underdog app"}],"claims":["Lab claim: 4.94% average word error rate on the Open ASR Leaderboard's seven English sets (7.82% on AMI meetings), against 4.94% for Parakeet TDT 0.6B v3 and 6.04% for Whisper large-v3. Underdog scored Bark itself with the leaderboard's code; it is not on the leaderboard.","Lab claim: the #1 frontier speech model, and transcribes a minute of audio in under 0.25 s on a MacBook. The model card gives about 1.5 s per minute on an M2 laptop with stock mlx-audio."],"sources":[{"kind":"lab","title":"Underdog announcement on X","url":"https://x.com/UnderdogAI/status/2112733007484518470","date":"2026-10-20"},{"kind":"lab","title":"Underdog Bark 3.0 page","url":"https://underdog.ai/bark","date":"2026-10-20"},{"kind":"repo","title":"Underdog-Bark-3.0 on Hugging Face","url":"https://huggingface.co/ConwayResearch/Underdog-Bark-3.0","date":"2026-10-20"}],"firstBenchmark":null},"organization":{"id":"underdog-ai","name":"Underdog","slug":"underdog-ai","country":null,"url":"https://underdog.ai/bark","color":"#11110F","mark":"U"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Oct 20, 2026; no independent results yet.","lead":"As of Oct 22, 2026, Underdog Bark 3.0 is pending benchmarks on isitsota: announced Oct 20, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}