Connected models in Desktop still create separate semantic models when published?

Hi Avih 

Yes, what you're experiencing is the expected behavior in Power BI. When you publish a report from Power BI Desktop—even if it connects to another semantic model (dataset) in Desktop—Power BI Service treats the published file as a standalone report and automatically creates a new semantic model associated with it. This happens because Power BI Desktop doesn’t truly support live connections to Power BI Service datasets; it connects to local copies of those datasets during development. When published, these connections don’t carry over as live links but are instead embedded as either static imports or re-published semantic models, resulting in duplication.

To avoid creating duplicate semantic models in the service, the best practice is to develop and maintain your centralized or “main” semantic model separately—publish it once to the Power BI Service—and then, for all dependent reports, do not build them in Desktop by connecting to a local version of that semantic model. Instead, after publishing the main model, create the dependent reports directly in the Power BI Service or in Power BI Desktop by using the “Power BI datasets” connector, which connects live to the published dataset in the service. This way, when you publish those reports, you can choose to publish just the report (without its own dataset), thereby preserving a single source of truth and avoiding dataset duplication. You can do this by publishing a report file created from a live connection (which shows "Live connection" mode in Desktop) to the same workspace as the main dataset or a different one, depending on your governance setup.