Parakeet-TDT 0.6B
The open ASR speed king — an hour of audio transcribed in about a second
About
Parakeet-TDT 0.6B is NVIDIA's open speech recognition model built on a FastConformer encoder with a TDT (token-and-duration transducer) decoder. Version 2 topped the Hugging Face Open ASR leaderboard at 6.05% average word error rate for English, and version 3 extends to 25 European languages with automatic language ID at 6.34% average WER, with built-in punctuation, capitalization, and word timestamps. Its headline number is speed: a real-time factor around 3386, meaning an hour of audio transcribes in roughly a second on server GPUs. Weights are CC-BY-4.0, so commercial use is allowed with attribution. The honest weakness is vocabulary: NVIDIA's own model card warns that words outside the training vocabulary are unlikely to be recognized, and there is no hotword-boosting API in the open checkpoints.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team/community-reported: 6.05% avg WER on the HF Open ASR leaderboard (v2, ranked #1); 6.34% avg WER (v3); RTFx ~3386 (NVIDIA)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: HF model card nvidia/parakeet-tdt-0.6b-v3 verified with full benchmark table (6.34% avg WER); CC-BY-4.0 commercial terms and RTF ~3386 verified across sources; v3 exact release day not verified — releasedAt null; vocabulary-biasing weakness noted.