Parakeet RNNT 0.6b
NVIDIA · released Dec 28, 2023 · nvidia/parakeet-rnnt-0.6b
Speech to textTranscribes a recording into words
- Type
- Open weightsCreative Commons Attribution 4.0
- Languages
- 1
- Size
- 0.6B
Context measured in tokens
Our take
Written Sep 11, 2026Parakeet RNNT is a 0.6-billion-parameter speech-to-text model from NVIDIA built for extreme speed on clean audio. It processes an hour of English speech in under a second, but its accuracy collapses on meetings, accents and everyday video.
Pick this for batch transcription of clean, read-aloud English where throughput is everything and you can tolerate about one word in sixteen wrong on average. Use it in resource-constrained or edge environments where model size is the binding constraint, provided you can meet the attribution requirement. Skip it if your audio includes crosstalk, accents, or background noise, or if you need a hosted API rather than self-hosting.
The case for it
- Extreme benchmark speed: 5,431 times real time on the leaderboard harness.
- Strong on clean read-aloud audio, with a word error rate better than most of 78 models.
- 0.6 billion parameters, small enough for embedded or edge deployment.
The case against it
- Error rate rises more than elevenfold on meeting-room audio with crosstalk.
- Worse than most on podcasts, video and accented speech.
- No commercial hosting options listed; you must run it yourself.
How good is it?
An open speech-to-text model for turning recordings into written text, though it trails most models on meetings and accented speech.
- turning spoken English into written textOpen ASR WER · 58th of 76
- transcribing recordings of meetings in a roomRecorded meetings · 80th of 92
- transcribing speakers with a range of accentsAccented speech · 63rd of 76
TranscriptionTurning speech into text2.5 of 5Open ASR WER · 58th of 76
93.8%
Misses roughly one word in 16, averaged over nine English test sets.
5,431×12th of 74
an hour of audio in under a second, on the board's own hardware. Your machine will differ.
1
Listed on the model card. The accuracy above is English only.
Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case; the placing beneath each rate is against every model measured on that set.
The figures above come from the Open ASR Leaderboard, an independent public test that runs every model on the same recordings. It is the only measurement of transcription quality we know of, so there are no other scores to show.
Each of these is the same transcription job on a different kind of recording, so together they say where it holds up and where it slips — not how closely it follows an instruction.
Every published score for this model9 scoresEvery figure we hold, from 9 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
Comfortable fit
GeForce RTX 4090 · 24 GB
Room to spare. 21.2 GB spare means a 10% error in the size would not change the answer.
Apple M1 Pro (16-core GPU) · 32 GB
Room to spare. 22.4 GB spare means a 10% error in the size would not change the answer.
Apple M1 (8-core GPU, 8GB unified) · 8 GB
Room to spare. 4.4 GB spare means a 10% error in the size would not change the answer.
These cards answer whether Parakeet RNNT 0.6b loads, not how fast it transcribes. Throughput figures for a transcription model come from its text decoder, so treat this as a fit answer rather than a speed one.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first. This is a fit answer: whether it loads, not how fast it transcribes.
Models people weigh against Parakeet RNNT 0.6b
When we formed this view
Recent changes
What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
Licence and identifiers
What the licence allowsCreative Commons Attribution 4.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Creative Commons Attribution 4.0
Permissive content license: any use with attribution. Common for datasets and some model weights.
Identifiers
- Hugging Face
- nvidia/parakeet-rnnt-0.6b
- Architecture
- Dense
- Takes in, gives back
- Audio in, text out
- Catalogue slug
- nvidia-parakeet-rnnt-0-6b