Whisper Model Guide
Whisper Model Sizes Explained: Tiny, Base, Small, Medium, Large for Linux
Picking the right model is the biggest performance decision for Linux voice dictation. Use this guide to choose between tiny, base, small, medium, and large based on RAM, speed, and real-world transcription accuracy in Vocalinux.
| Model | Size | RAM (CPU) | RAM (GPU) | Accuracy | Speed | Best For |
|---|---|---|---|---|---|---|
| Tiny | ~74MB | ~0.8-1.2GB | ~0.5-0.8GB VRAM | Good | Fastest | Older laptops, lowest latency, quick notes |
| Base | ~141MB | ~1.2-1.8GB | ~0.8-1.2GB VRAM | Good+ | Very fast | Balanced real-time dictation on most systems |
| Small | ~465MB | ~2-3GB | ~1.5-2.5GB VRAM | High | Fast | Most users wanting stronger punctuation and names |
| Medium | ~1.5GB | ~4-6GB | ~3-5GB VRAM | Very high | Moderate | Accuracy-focused dictation, technical writing |
| Large | ~3.0GB | ~8GB+ | ~6-10GB VRAM | Best | Slowest | Maximum quality with strong GPU hardware |
Which model should you choose?
Tiny or Base
- Best if you care about low latency over absolute transcription quality.
- Great for older CPUs, battery-conscious laptops, and quick chat replies.
Small (recommended default)
- Best balance for most Linux dictation workflows.
- Better punctuation and proper nouns while staying near real-time.
Medium
- Use when accuracy is more important than response time.
- Ideal for technical dictation, multilingual speech, and noisy environments.
Large
- Choose only if you have strong GPU hardware and want maximum quality.
- Best for polished long-form writing where errors are costly.
Need more speed?
Enable GPU acceleration to run larger models faster, especially medium and large.
GPU acceleration guideComparing engines?
See whisper.cpp vs Whisper vs VOSK before deciding your full dictation stack.
Engine comparisonReady to install?
Use the install guide to get Vocalinux running with the model size that fits your hardware.
Installation guideInstall Vocalinux and start dictating today
Start with small if unsure, then move up or down after a day of use. Vocalinux makes model changes simple, so you can tune for your exact speed/accuracy preference.
Install Vocalinux