Whisper large-v3
A solid pick for self-hosting and full data control.
Overview
Whisper large-v3 is an encoder-decoder model for speech recognition and translation, trained on millions of hours of weakly supervised audio, which gives it solid multilingual zero-shot robustness. Its open weights under Apache 2.0 and laptop-tier hardware requirements make it a practical pick for on-premise deployment in Europe, keeping audio data on local infrastructure. The real constraint: a 30-second receptive field means longer recordings need chunking or sequential decoding, trading speed against accuracy.
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
- Meeting transcription
- Multilingual subtitling
- Speech-to-text translation
- On-premise voice assistants
- Audio content archiving
Access & availability
Key specifications
Privacy
Run it locally
| Quantization | Disk | RAM / VRAM | Typical hardware |
|---|---|---|---|
| Q4 · recommended | 0.9 GB | 3 GB | Any recent PC/Mac |
| Q8 · balanced | 1.7 GB | 4 GB | Any recent PC/Mac |
| FP16 · max quality | 3.1 GB | 5 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 openai/whisper-large-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 | — |