Closed Model vs Open-Weight Model
People think a closed model is better and an open one is worse, and that is not the split. The split is who holds the model file. With a closed model, the maker keeps it, and you reach it online. With an open-weight model, you can download it and run it on your own computer. Both kinds come in strong and weak.
Closed Model
An AI model you can only use online. The maker keeps the model itself.
A closed model is an AI model whose weights the maker keeps to itself. You can only use it through the maker's app or its paid service online. You cannot download it, run it on your own computer, or change how it was trained. The maker can update it, change its price or switch it off. Most frontier models are closed, though some labs share open models as well.
Open-Weight Model
A model whose weights you can download and run yourself.
An open-weight model is a model whose trained weights are published. Anyone can download it and run it on their own machine. You can use it offline, keep your data private and train it further. Closed models stay on the maker's servers, and you reach them online. Open weights are not open source: the data and code are often still private, and the licence may limit use.
In one line
- Closed Model
- An AI model you can only use online. The maker keeps the model itself.
- Open-Weight Model
- A model whose weights you can download and run yourself.