Open vs closed weights

Some models you can only reach through a company's app. Others you can download and keep forever. What's the real difference — and when does each one win?

Rent a room, or own a house

A closed model lives on the company's servers; you rent access through their app or API. A model with open weights is different — the whole trained model is published as a file you can download, run on your own machine, and even modify. Closed is a room you rent; open is a house you own.

Which one wins?

Closed models are the easy path: nothing to set up, always the newest version, no hardware needed. Open weights trade that convenience for control — run it offline, keep your data private, tune it to your task, pay no ongoing bill. Neither is 'better'; they answer different needs. Most people start closed; builders and the privacy-conscious reach for open.

The short version

• —Closed models are rented through an app or API — easy, always current.
• —Open-weight models are downloaded and owned — private, tunable, offline.
• —Pick by need: convenience (closed) vs control (open).