Parakeet TDT 0.6B v3
A solid pick for self-hosting and full data control.
Overview
Parakeet TDT 0.6B v3 is NVIDIA's compact ASR model, small enough (0.6B parameters) to run on laptop-class hardware. It auto-detects and transcribes 25 European languages with punctuation and timestamps, making it a practical base for multilingual transcription or subtitling tools. Quality isn't uniform across languages, and NVIDIA flags a specific gap on Portuguese, trained on the European variant but often benchmarked against Brazilian Portuguese. Open weights under CC-BY-4.0 mean that, self-hosted, all audio and text stay on the user's own infrastructure regardless of the vendor's US origin.
Skill profile
Not disclosed
Strengths
- Open-source and self-hostable
Limitations
- API pricing not disclosed
Who is it for
- you want to control cost or self-host
- you have GDPR constraints
Ideal use cases
- Multilingual meeting transcription
- Automated video subtitling
- On-premise voice assistants
- Confidential call analytics
Access & availability
Key specifications
Privacy
Run it locally
| Quantization | Disk | RAM / VRAM | Typical hardware |
|---|---|---|---|
| Q4 · recommended | 0.3 GB | 2 GB | Any recent PC/Mac |
| Q8 · balanced | 0.6 GB | 3 GB | Any recent PC/Mac |
| FP16 · max quality | 1.2 GB | 3 GB | Any recent PC/Mac |
Estimates for a moderate context. Long contexts need more RAM (KV cache).
Deploy
Copy-ready commands generated from this card. Adjust context length and GPU count to your hardware.
OpenAI-compatible server for production on NVIDIA GPUs.
pip install vllm
vllm serve nvidia/parakeet-tdt-0.6b-v3 \
--max-model-len 32768 \
--tensor-parallel-size 1 \
--dtype autoAdvanced data · for expertsArchitecture, modalities, detailed cost, full benchmarks▾
| Arena Elo | — |
| MMLU | — |
| GPQA | — |
| HumanEval | — |
| SWE-Bench | — |
| MATH | — |