Lesson 3 of 7
Closed vs open weights
6 min read
You can use most famous AIs, but you can't have them. A few, though, you can download and keep forever. What's the difference — and when does it matter?
Rent a room, or own the house
A closed model lives on the company's computers. You reach it through their app or an API — you rent access. An [open-weights](glossary://open-weights) model publishes its trained insides as a file you can download, run, and keep. Renting a room versus owning the house: one you just walk into, the other is yours to renovate.
Closed means easiest and always current, but on their terms. Open means it's yours to run offline and modify — at the cost of doing the hosting yourself.
Which one, when?
For most people most of the time, closed wins: nothing to install, always the newest, top quality. Open comes into its own when you need privacy (nothing leaves your machine), control (change or fine-tune it), or no ongoing bill. The frontier is usually closed; open weights trail a little behind, but close the gap fast.
Reach for open weights when privacy, control, or cost matter more than having the absolute latest and greatest.
'Open weights' isn't the same as fully open-source. You usually get the finished model to run, not the data or recipe used to make it — and some licences limit commercial use. Read the licence before you build on one.
The shape of it
- —Closed models are rented through an app or API; open-weights models you download and keep.
- —Closed is easiest and most current; open gives privacy, control, and no ongoing bill.
- —Open weights isn't full open-source — check what the licence actually allows.
A clinic wants an AI that keeps patient notes entirely on its own computers, never sending them to anyone. Which fits?
Continue in the app
Take the whole Which AI should I use? course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.