An open-source AI model is one whose trained parameters, called weights, are published for anyone to download, run and often modify, subject to a licence. Many are technically "open-weight": the weights are public, but the training data and code may not be. Proprietary models, by contrast, are accessed only through the owner's service.
To use an open model you run it on your own servers or rent hosting for it. That gives control over data location and customisation, but you or your IT partner must handle setup, security, updates, monitoring and capacity.
For retailers and hospitality groups, open models can suit privacy-sensitive tasks, offline or in-store use, or high-volume simple jobs where a small model on your own hardware is cheaper than paying per use. Proprietary models usually lead on the hardest tasks and are far easier to start with.
Read the licence carefully, since some restrict commercial use or large-scale deployment. Compare quality on your own tasks, not just published rankings, and count the real cost of running and supporting it. Security patching and safety filtering become your responsibility.
The open versus proprietary choice is about control against convenience, and many businesses end up using a mix of both.