Parakeet-TDT-0.6B-v2
CC-BY-4.0 open 600M ASR model transcribing an hour of audio in a second.
About
NVIDIA's Parakeet-TDT-0.6B-v2 is a 600M-parameter FastConformer-TDT speech recognition model under CC-BY-4.0 — commercially usable, genuinely open. Its headline stat is speed: an RTFx of 3380, transcribing an hour of audio in about a second on GPU, topping the Hugging Face Open ASR Leaderboard at 6.05% WER when released.
It is English-only and shines on clean audio; accented or noisy speech degrades it like any ASR model. A newer multilingual v3 exists covering 25 European languages if you need non-English.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- 6.05% WER on HF Open ASR Leaderboard (reported)
- RTFx 3380 — an hour of audio in ~1 second on GPU (reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Official NVIDIA HF page; CC-BY-4.0 — commercially usable; Released 2025-05-01; 600M params; 6.05% WER on Open ASR Leaderboard at release; English-only; newer multilingual v3 exists for non-English needs. License: CC-BY-4.0 confirmed from official Hugging Face model card..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Official NVIDIA HF page; CC-BY-4.0 — commercially usable
- · Released 2025-05-01; 600M params; 6.05% WER on Open ASR Leaderboard at release
- · English-only; newer multilingual v3 exists for non-English needs