In AI services, an endpoint is an address or deployment target through which an application sends inference requests. It may expose a particular model, version, provider configuration or region, even when another endpoint uses the same underlying model family.
Two endpoints serving nominally the same model can differ in system instructions, sampling settings, input limits, quantization, routing, tool availability or provider updates. These choices may affect cost, latency, reliability and the output users receive.
An endpoint is also an operational boundary. Teams may configure authentication, rate limits, geographic processing, traffic logs and data retention at the service level. A public model identifier alone does not establish what data rules apply to the endpoint hosting it.
For repeatable open-versus-closed model tests, record the endpoint, provider, model version, relevant parameters and test date. Recheck the behavior after configuration changes or service updates.