The largest language models can only be reached through the provider's service, often in the US. An open model, on the other hand, can run in your own environment, so that the data never leaves the building. Examples are models for transcription, search and smaller language models.
The price is capacity. An open model needs a GPU and the largest models need several. That is why we use open models for well-defined tasks and a European provider or the customer's own GPU for the big ones.