Can Your Computer Run Local AI? Whisper Check
Tell us your RAM, graphics card and system, and we’ll tell you which Whisper model you can run locally, roughly how fast, and whether it beats the cloud for you.
Which model to choose, minus the marketing
Model size decides everything: quality, speed and whether your computer can cope. Comparing «local AI» with «cloud AI» without saying which model is used is cheating, because it isn’t the same tool.
| Model | Memory | Quality | When |
|---|---|---|---|
| tiny | 1 GB | Low | Good for searching inside audio, not for publishing. |
| base | 1.5 GB | Low–medium | Struggles with proper names and strong accents. |
| small | 2.5 GB | Medium | The sensible balance without a dedicated graphics card. |
| medium | 5 GB | High | A noticeable jump. Slow on CPU alone. |
| large-v3 turbo | 8 GB | Best in class | Paid-service quality, on your own machine. |
When local really wins
- • Sensitive material: meetings with client data, interviews under NDA. It’s not that they promise not to look: it never leaves your machine.
- • High volume: services charge by the minute. A whole archive is a budget decision in the cloud and a weekend of machine time locally.
- • No connection: a shoot with no signal, a flight, hotel wifi.
- • Ongoing use: if you transcribe every week, the subscription never stops running. A downloaded model doesn’t expire.
And when the cloud still wins
- • If your computer is modest and you need high quality: a small model makes more mistakes than a paid service.
- • If you need to tell speakers apart accurately, with many people talking: commercial services are still ahead there.
- • If you transcribe very little: for three recordings a year, it isn’t even worth downloading the model.