diff --git a/src/lightning/pytorch/cli.py b/src/lightning/pytorch/cli.py
index 26af335f7be93..9f970f0c22a42 100644
--- a/src/lightning/pytorch/cli.py
+++ b/src/lightning/pytorch/cli.py
@@ -138,7 +138,10 @@ def add_lightning_class_args(
             if issubclass(lightning_class, Callback):
                 self.callback_keys.append(nested_key)
             if subclass_mode:
+                print("Subclass mode")
+                print(nested_key)
                 return self.add_subclass_arguments(lightning_class, nested_key, fail_untyped=False, required=required)
+            print(nested_key)
             return self.add_class_arguments(
                 lightning_class,
                 nested_key,
diff --git a/tests/tests_pytorch/test_cli.py b/tests/tests_pytorch/test_cli.py
index 56b58d4d157a1..0a3b885211f78 100644
--- a/tests/tests_pytorch/test_cli.py
+++ b/tests/tests_pytorch/test_cli.py
@@ -597,6 +597,53 @@ def add_arguments_to_parser(self, parser):
     assert cli.model.num_classes == 5
 
 
+# Notes:
+# - if variable is class attribute, it will only be present after instantiation -> apply_on="instantiate"
+# - if you link argumetns and then try to pass them additionally, it will raise an error
+# - if you initilize the model from the cmd instead of passing it as an instance to CLI init_args needs to be
+#   added additionally
+# - if you initilize the model from the cmd instead of passing it as an instance to CLI apply on instantiate needs to
+#   be always true for linked arguments
+
+
+def test_lightning_cli_link_arguments_init():
+    # Will not work without init_args ("--data.init_args.batch_size=12")
+    class MyLightningCLI(LightningCLI):
+        def add_arguments_to_parser(self, parser):
+            parser.link_arguments("data.batch_size", "model.batch_size")
+            parser.link_arguments("data.num_classes", "model.num_classes", apply_on="instantiate")
+
+    cli_args = [
+        "--data.batch_size=12",
+    ]
+
+    with mock.patch("sys.argv", ["any.py"] + cli_args):
+        cli = MyLightningCLI(BoringModelRequiredClasses, BoringDataModuleBatchSizeAndClasses, run=False)
+
+    assert cli.model.batch_size == 12
+    assert cli.datamodule.batch_size == 12
+    assert cli.model.num_classes == 5
+    assert cli.datamodule.num_classes == 5
+
+    # Will work without init_args ("--data.batch_size=12")
+    class MyLightningCLI(LightningCLI):
+        def add_arguments_to_parser(self, parser):
+            parser.link_arguments("data.batch_size", "model.init_args.batch_size", apply_on="instantiate")
+            parser.link_arguments("data.num_classes", "model.init_args.num_classes", apply_on="instantiate")
+
+    cli_args = [
+        "--data=tests_pytorch.test_cli.BoringDataModuleBatchSizeAndClasses",
+        "--model=tests_pytorch.test_cli.BoringModelRequiredClasses",
+        "--data.batch_size=12",
+    ]
+
+    with mock.patch("sys.argv", ["any.py"] + cli_args):
+        cli = MyLightningCLI(run=False)
+
+    assert cli.datamodule.batch_size == 12
+    assert cli.model.batch_size == 12
+
+
 class EarlyExitTestModel(BoringModel):
     def on_fit_start(self):
         raise MisconfigurationException("Error on fit start")