Whisper Small
244M-parameter Whisper. The smallest Whisper checkpoint considered 'production grade' for non-English audio. Sweet spot for laptops with iGPU/Metal or modest discrete GPUs.
Pick this when base is not accurate enough on your target language and large is too heavy. Below 'edge' threshold strictly by param count, but realistically a consumer-laptop model.
Overview
244M-parameter Whisper. The smallest Whisper checkpoint considered 'production grade' for non-English audio. Sweet spot for laptops with iGPU/Metal or modest discrete GPUs.
Strengths
- Materially better non-English WER than tiny/base
- Realtime on Apple Silicon with whisper.cpp Metal backend
- Apache-2.0; widely fine-tuned (medical, legal, domain-specific forks)
- Strong punctuation and casing out of the box
Weaknesses
- Still hallucination-prone on silence and music
- Outclassed on WER by distil-large-v3 at similar throughput on GPU
- Slower than Parakeet on NVIDIA hardware
- 30s window — long-form needs WhisperX or similar
Quantization variants
Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.
| Quantization | File size | VRAM required |
|---|---|---|
| Q4_K_M | 0.1 GB | 1 GB |
Get the model
HuggingFace
Original weights
Source repository — direct quantization required.
Hardware that runs this
Cards with enough VRAM for at least one quantization of Whisper Small.
Models worth comparing
Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.
Frequently asked
What's the minimum VRAM to run Whisper Small?
Can I use Whisper Small commercially?
What's the context length of Whisper Small?
Source: huggingface.co/openai/whisper-small
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.
Related — keep moving
Verify Whisper Small runs on your specific hardware before committing money.