GLOSSARY
Open weight
Open model weights are accessible learned parameters of an AI model. They may enable self-hosting or certain modifications. Actual rights to use, modify and redistribute depend on the applicable terms and on which supporting components are available.
Model access does not define the operating arrangement
A sales team wants to process confidential proposal documents in an environment under its own responsibility. A model with accessible weights could be an option. It still needs suitable computing resources, an application, controlled access and a process for updates.
The decision transfers work from an external service provider to the organisation's own operation. That can create control while also consuming specialist time and support capacity. With irregular usage, the cost of reserved infrastructure should be compared with the actual workload. Procurement and IT need a shared view of those obligations before agreeing on a deployment.
Check rights and quality in context
Start with the licence and representative documents. Examine substantive errors, response times, peak demand and the effort of later model updates. The distinction from open-source AI prevents assumptions about rights based only on available parameters. Self-hosting also requires named owners and a credible maintenance plan.