What causes MODEL_VERSION_NOT_FOUND?

The model version was deleted from the MLflow model registry or the Unity Catalog model. The job references a hard-coded version number that was archived or replaced by a new promotion workflow. The workspace is using the workspace model registry but the model was registered in Unity Catalog (or vice versa). The model name or version alias used in code does not match the registered name

How do I fix MODEL_VERSION_NOT_FOUND?

Step 1: In the Databricks UI, open the Catalog (Unity Catalog) or Models (workspace registry) and confirm the model name and version exist.. Step 2: If the version was deleted, determine the correct latest version and update the pipeline or serving endpoint configuration.. Step 3: Replace hard-coded version numbers with aliases such as 'champion' or 'production' to avoid breakage when models are promoted.. Step 4: Verify that the code and the serving endpoint are reading from the same registry (Unity Catalog three-part name vs. workspace registry name).. Step 5: Re-run the job or reload the serving endpoint after correcting the model reference.

High severityconfiguration

Power BI Refresh Error:
MODEL_VERSION_NOT_FOUND

What does this error mean?

An MLflow or Unity Catalog operation referenced a registered model version that does not exist, has been deleted, or is in a different model registry than the one being queried.

Common causes

1The model version was deleted from the MLflow model registry or the Unity Catalog model
2The job references a hard-coded version number that was archived or replaced by a new promotion workflow
3The workspace is using the workspace model registry but the model was registered in Unity Catalog (or vice versa)
4The model name or version alias used in code does not match the registered name

How to fix it

1Step 1: In the Databricks UI, open the Catalog (Unity Catalog) or Models (workspace registry) and confirm the model name and version exist.
2Step 2: If the version was deleted, determine the correct latest version and update the pipeline or serving endpoint configuration.
3Step 3: Replace hard-coded version numbers with aliases such as 'champion' or 'production' to avoid breakage when models are promoted.
4Step 4: Verify that the code and the serving endpoint are reading from the same registry (Unity Catalog three-part name vs. workspace registry name).
5Step 5: Re-run the job or reload the serving endpoint after correcting the model reference.

Frequently asked questions

What is the difference between model stages (Staging, Production) and model aliases?

Model stages are a legacy concept from the workspace MLflow registry. Unity Catalog model registry replaces stages with aliases, which are custom string labels you assign to a version. Databricks recommends migrating to Unity Catalog model registry and using aliases.

How do I prevent version number drift between environments?

Register models in Unity Catalog with a consistent naming scheme and use environment-specific aliases (e.g., dev-champion, prod-champion) mapped to version numbers. Pipelines reference the alias, not the version number.