Open vs. closed models
Who gets to download, modify and run the model?
The analogy
A closed model is like a restaurant: you eat the dish, but the recipe is secret and it's only served there. An open model is like a published recipe: anyone can cook it at home, tweak it to their taste, or open their own place with it.
In detail
With open-weights models, the parameters are published and anyone can download them, run them on their own infrastructure or fine-tune them. Closed models are only used through the provider's API. Open models give you control and privacy in exchange for managing the infrastructure; closed ones usually offer top capability with zero operational effort.
An example
A hospital that can't let data leave its servers runs an open model locally to summarize medical records; a startup that wants the best quality without managing GPUs uses a closed model's API.