Lesson 7 of 7
Free, paid, or local?
6 min read
You can even run capable AI on your own laptop for 'free'. So is local always cheapest — or is that a trap at low volume?
Three cost shapes
Each option has a different shape. A subscription is a flat line — same cost whatever you do. The [API](glossary://api) starts near zero and rises with every call. Local / open models are the opposite: a big upfront cost (hardware, setup, electricity) and then almost nothing per use. Plotted against volume, these lines cross.
Flat, rising, and upfront-then-flat — three shapes that cross as volume grows.
Find your break-even
For light or occasional use, the flat subscription or the metered API wins, and local's upfront cost never pays off. Local only becomes cheaper past a high, steady volume — plus it buys privacy, offline use, and no per-call fee. Don't guess: estimate your real monthly volume (last lesson's formula) and see which line is lowest there.
Local pays off only past a high, steady volume — below it, hosted wins.
Pick the option that's cheapest at your volume today, and re-check when your usage changes — the winner isn't fixed.
Choosing your setup
- —Subscription = flat; API = rises with use; local = big upfront, cheap per use.
- —Light use → hosted; heavy steady use → maybe local.
- —Estimate your volume, then pick the lowest line.
When does running a model locally actually save money?
Continue in the app
Take the whole What AI Costs course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.