Transcription · NVIDIA
Parakeet
Sub Rosa's default transcription: fast and accurate for everyday dictation and meetings, among the cheapest, in private mode.
Private Low costWhat sets it apart
Parakeet is a small 600M open model from NVIDIA that transcribes 25 European languages with automatic language detection, punctuation and word timestamps. It is built to be fast and cheap, with accuracy that trails the best paid services.
- 600M parameters, CC-BY-4.0
- 25 European languages, auto detection
- Word timestamps and punctuation built in
Recommended for: transcribe meetings and dictations.
Strengths and limits
Where it shines
- Default in Sub Rosa for dictation and meetings
- Cheapest private option, tied with Whisper
- Detects the language and adds punctuation
Think twice when
- Covers 25 European languages only: for others, pick Whisper
- 25 European languages, French and English included
Use cases
Dictate notes in French or English
Transcribe this two-minute voice memo with punctuation and capital letters. The language may switch between French and English.
It detects the language itself and punctuates by default.
Transcribe a long meeting cheaply
Transcribe this 90-minute team meeting with a timestamp on every sentence.
It handles up to 3 hours of audio and is among the cheapest options.
Against its rivals
- WhisperWhisper covers 99 languages at the same price, where Parakeet covers 25.#3 of 4 on Artificial Analysis word error rate (4.1%)
- ScribeScribe is more accurate on the Artificial Analysis test but costs more and is anonymized.#1 of 4 on Artificial Analysis word error rate (2.2%)
- xAI Speech to TextxAI is cheaper and more accurate on that test, but anonymized.#2 of 4 on Artificial Analysis word error rate (4%)
Version history
Each version, dated, with what it changed over the one before.
- Parakeet TDT 0.6B v2Earlier, not in Sub Rosa
- English-only FastConformer-TDT model with punctuation and timestamps
- Parakeet TDT 0.6B v3Latest
- Expands from English only to 25 European languages
- Automatic language detection and up to 3 hours of audio with local attention
Benchmarks
How to read these scores: what a benchmark tells you.
Announced by the maker
Measured by the maker with its own settings: useful, but not directly comparable with the independent scores above.
| Benchmark | Model | Score | Source |
|---|---|---|---|
| Artificial Analysis word error rate | Parakeet ASR | 6.3% | huggingface.co |
The model in this family
| Model | Use | Privacy | Price |
|---|---|---|---|
Parakeet ASRnvidia/parakeet-tdt-0.6b-v3 | Transcription | Private | $0.162 per hour of audio |
Specifications
| Specification | Parakeet TDT 0.6B v3 |
|---|---|
| Released | August 2025 |
| Parameters | 0.6B |
| Weights | Open, CC-BY-4.0 |
| Accepts | audio |
| Produces | text |
| Languages | 25 European languages |
| Price | $0.162 per hour of audio |
| Privacy | Private |
Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.
Sources
Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.