Delete a Version

Move a dataset version to Trash, where it is kept for 30 days before being permanently removed.

Deleting a dataset version moves it to your workspace Trash for 30 days before it is permanently removed. You can restore it from the Trash any time during the retention window.

Any in-flight training for the version is cancelled automatically when you move it to Trash. The trained model for the deleted version stops serving until the version is restored.

If you have hit your generated image quota, moving a version to Trash does not immediately free up your usage. Generated images are reclaimed when the version is permanently deleted (either after the 30-day retention window or manually from the Trash).

Delete a version

1. Open the project's Versions page and select the version you want to delete.

2. Click "Edit Dataset" and then "Delete Version".

3. Type the version name to confirm.

4. Click "Move to Trash".

Restore or permanently delete

See Trash for how to restore a deleted version or remove one immediately without waiting for the 30-day retention window. The version's parent project must be active (not itself in Trash) to restore.

Delete via API, SDK, or CLI

You can also move a version to Trash through:

• REST API: DELETE /:workspace/:project/:version

• Python SDK: version.delete()

• CLI: roboflow version delete <workspace>/<project>/<version>

To restore outside the web app use version.restore() from the SDK or roboflow version restore <workspace>/<project>/<version> from the CLI.