Back to Gap Map Model components / Speech & audio

Parakeet

open weights / Overall score: 4.0(strong)

NVIDIA's FastConformer transducer and CTC speech recognition models. The current Parakeet TDT 0.6B v3 transcribes 25 European languages with word-level timestamps and runs thousands of times faster than real time on a GPU; earlier checkpoints are English-only.

Most checkpoints are CC-BY-4.0; five newer ones, parakeet-unified-en and the multitalker streaming model among them, use the NVIDIA Open Model License, which sits in the same license tier. Trained and served with NeMo Speech.

Openness

3 medium confidence

3.0

weights: open(.nemo and safetensors checkpoints, ungated)
data: partial(Granary is released
code: open(NeMo training script and TDT configuration linked from the card)
license: CC-BY-4.0(weights)+NVIDIA-Open-Model-License(newer checkpoints)

Parakeet's weights are free to download under CC-BY-4.0, which asks only for attribution, and NVIDIA links the exact NeMo training script and configuration. Its pseudo-labeled Granary data is public, but the human-transcribed part of the mixture is NVIDIA's in-house NeMo ASR Set 3.0, so the model cannot be fully reproduced from what is released.

https://huggingface.co/api/models?author=nvidia&search=parakeet recorded 2026-09-27
Family listing: most repositories tagged "license:cc-by-4.0"; parakeet-unified-en-0.6b and multitalker-parakeet-streaming-0.6b-v1 are tagged "license:other" (the NVIDIA Open Model License).
https://huggingface.co/api/models/nvidia/parakeet-tdt-0.6b-v3 recorded 2026-09-27
"license:cc-by-4.0" tag, "gated":false, with parakeet-tdt-0.6b-v3.nemo in the file list.
https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3/raw/main/README.md recorded 2026-09-27
"The model was trained on the combination of [Granary dataset's ASR subset](https://huggingface.co/datasets/nvidia/Granary) and in-house dataset NeMo ASR Set 3.0"; "Training was conducted using this [example script]"; "GOVERNING TERMS: Use of this model is governed by the [CC-BY-4.0]".

Adoption

4 high confidence

4.0

Hugging Face downloads summed over the three declared checkpoints, which carry most of the family's use: led by the v3 model. Downloads count file fetches, not transcriptions.

https://huggingface.co/api/models?author=nvidia&search=parakeet recorded 2026-09-27
Trailing-30-day downloads: parakeet-tdt-0.6b-v2 193,528 and parakeet-ctc-1.1b 372,919.
https://huggingface.co/api/models/nvidia/parakeet-tdt-0.6b-v3 recorded 2026-09-27
"downloads":563206 for nvidia/parakeet-tdt-0.6b-v3.

Capability

4 medium confidence

4.0

Parakeet is the throughput leader among accurate open recognizers, a fraction of a point behind the top of the board while running many times faster than larger models.

https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/resolve/d2c5b384deccdb82834f41aeaffcc618c00efa2f/english_short_latest.csv recorded 2026-09-27
Results CSV rows "nvidia/parakeet-tdt-0.6b-v2,4.7025,6024.67" and "nvidia/parakeet-tdt-0.6b-v3,4.85875,6076.07".

Verified 2026-09-27