diff --git a/optimum/intel/openvino/modeling_visual_language.py b/optimum/intel/openvino/modeling_visual_language.py
index e0c203d8fe..76e4320fb4 100644
--- a/optimum/intel/openvino/modeling_visual_language.py
+++ b/optimum/intel/openvino/modeling_visual_language.py
@@ -578,6 +578,11 @@ def half(self):
                 compress_model_transformation(model)
         return self
 
+    def to(self, device):
+        self.language_model.to(device)
+        super().to(device)
+        return self
+
     def forward(
         self,
         input_ids,
diff --git a/tests/openvino/test_modeling.py b/tests/openvino/test_modeling.py
index 6c68438c73..c2efd12f0d 100644
--- a/tests/openvino/test_modeling.py
+++ b/tests/openvino/test_modeling.py
@@ -1944,7 +1944,7 @@ def test_compare_to_transformers(self, model_arch):
         inputs = self.gen_inputs(model_arch, "What is shown on this image?", self.IMAGE)
 
         ov_model = OVModelForVisualCausalLM.from_pretrained(
-            model_id, export=True, trust_remote_code=model_arch in self.REMOTE_CODE_MODELS
+            model_id, export=True, trust_remote_code=model_arch in self.REMOTE_CODE_MODELS, compile=False
         )
         self.assertIsInstance(ov_model, MODEL_TYPE_TO_CLS_MAPPING[ov_model.config.model_type])
         self.assertIsInstance(ov_model.vision_embeddings, OVVisionEmbedding)
@@ -1953,6 +1953,27 @@ def test_compare_to_transformers(self, model_arch):
             self.assertTrue(hasattr(ov_model, additional_part))
             self.assertIsInstance(getattr(ov_model, additional_part), MODEL_PARTS_CLS_MAPPING[additional_part])
         self.assertIsInstance(ov_model.config, PretrainedConfig)
+        ov_model.to("AUTO")
+        self.assertTrue("AUTO" in ov_model._device)
+        self.assertTrue("AUTO" in ov_model.vision_embeddings._device)
+        self.assertTrue(ov_model.vision_embeddings.request is None)
+        self.assertTrue("AUTO" in ov_model.language_model._device)
+        self.assertTrue(ov_model.language_model.request is None)
+        self.assertTrue(ov_model.language_model.text_emb_request is None)
+        for additional_part in ov_model.additional_parts:
+            self.assertTrue("AUTO" in getattr(ov_model, additional_part)._device)
+            self.assertTrue(getattr(ov_model, additional_part).request is None)
+        ov_model.to("CPU")
+        ov_model.compile()
+        self.assertTrue("CPU" in ov_model._device)
+        self.assertTrue("CPU" in ov_model.vision_embeddings._device)
+        self.assertTrue(ov_model.vision_embeddings.request is not None)
+        self.assertTrue("CPU" in ov_model.language_model._device)
+        self.assertTrue(ov_model.language_model.request is not None)
+        self.assertTrue(ov_model.language_model.text_emb_request is not None)
+        for additional_part in ov_model.additional_parts:
+            self.assertTrue("CPU" in getattr(ov_model, additional_part)._device)
+            self.assertTrue(getattr(ov_model, additional_part).request is not None)
         # pytorch minicpmv is not designed to be used via forward
         if "minicpmv" not in model_arch:
             set_seed(SEED)
@@ -2015,6 +2036,15 @@ def test_generate_utils(self, model_arch):
 
         gc.collect()
 
+    @parameterized.expand(SUPPORTED_ARCHITECTURES)
+    def test_model_can_be_loaded_after_saving(self, model_arch):
+        model_id = MODEL_NAMES[model_arch]
+        with TemporaryDirectory() as save_dir:
+            ov_model = OVModelForVisualCausalLM.from_pretrained(model_id, compile=False)
+            ov_model.save_pretrained(save_dir)
+            ov_restored_model = OVModelForVisualCausalLM.from_pretrained(save_dir, compile=False)
+            self.assertIsInstance(ov_restored_model, type(ov_model))
+
 
 class OVModelForSpeechSeq2SeqIntegrationTest(unittest.TestCase):
     SUPPORTED_ARCHITECTURES = ("whisper",)