diff --git a/litellm/llms/vertex_ai/gemini/transformation.py b/litellm/llms/vertex_ai/gemini/transformation.py
index e50954b8f9..18c6e54e8c 100644
--- a/litellm/llms/vertex_ai/gemini/transformation.py
+++ b/litellm/llms/vertex_ai/gemini/transformation.py
@@ -284,7 +284,7 @@ def _gemini_convert_messages_with_history(  # noqa: PLR0915
         raise e
 
 
-def _transform_request_body(
+def _transform_request_body( # noqa: PLR0915
     messages: List[AllMessageValues],
     model: str,
     optional_params: dict,
@@ -295,13 +295,35 @@ def _transform_request_body(
     """
     Common transformation logic across sync + async Gemini /generateContent calls.
     """
+    # Duplicate system message as user message for Gemini 
+    duplicate_system_as_user = optional_params.pop("duplicate_system_as_user_for_gemini", True)
+
+    # Check if all messages are system messages
+    all_system_messages = all(message["role"] == "system" for message in messages)
+
     # Separate system prompt from rest of message
     supports_system_message = get_supports_system_message(
         model=model, custom_llm_provider=custom_llm_provider
     )
-    system_instructions, messages = _transform_system_message(
-        supports_system_message=supports_system_message, messages=messages
-    )
+
+    system_instructions = None
+    # If all messages are system messages, add a user message to the end
+    if (all_system_messages and supports_system_message and messages):
+        # Always create system instruction
+        system_content = messages[0].get("content", "")
+        system_part = PartType(text=system_content) # type: ignore
+        system_instructions = SystemInstructions(parts=[system_part])
+        
+        # Only duplicate as user message if flag is set
+        if duplicate_system_as_user or litellm.modify_params:
+            user_message = cast(AllMessageValues, {
+                "role": "user",
+                "content": system_content
+            })
+            messages = [user_message]
+        else:
+            messages = []
+
     # Checks for 'response_schema' support - if passed in
     if "response_schema" in optional_params:
         supports_response_schema = get_supports_response_schema(
diff --git a/tests/litellm/llms/vertex_ai/gemini/test_gemini_transformation.py b/tests/litellm/llms/vertex_ai/gemini/test_gemini_transformation.py
new file mode 100644
index 0000000000..cb9cdee0be
--- /dev/null
+++ b/tests/litellm/llms/vertex_ai/gemini/test_gemini_transformation.py
@@ -0,0 +1,71 @@
+import litellm
+from litellm.llms.vertex_ai.gemini.transformation import _transform_request_body
+
+
+def test_system_message_conversion_gemini():
+    """Test that system-only messages are properly handled for Gemini"""
+    # Case 1: Default behavior - duplicate system as user
+    messages = [{"role": "system", "content": "You are a helpful assistant"}]
+    
+    # Create mock objects for the test
+    model = "gemini-2.0-flash"
+    custom_llm_provider = "gemini"
+    optional_params = {}
+    litellm_params = {}
+    
+    result = _transform_request_body(
+        messages=messages, # type: ignore
+        model=model,
+        optional_params=optional_params,
+        custom_llm_provider=custom_llm_provider,
+        litellm_params=litellm_params,
+        cached_content=None
+    )
+    
+    # Verify that contents has user message
+    assert len(result["contents"]) > 0
+    assert result["contents"][0]["role"] == "user" # type: ignore
+    assert "system_instruction" in result
+    
+    # Case 2: Disable duplication
+    optional_params = {"duplicate_system_as_user_for_gemini": False}
+    
+    # Save original modify_params value
+    original_modify_params = litellm.modify_params
+    litellm.modify_params = False
+    
+    result_no_duplicate = _transform_request_body(
+        messages=messages.copy(), # type: ignore
+        model=model,
+        optional_params=optional_params,
+        custom_llm_provider=custom_llm_provider,
+        litellm_params={},
+        cached_content=None
+    )
+    
+    # Restore original modify_params value
+    litellm.modify_params = original_modify_params
+    
+    # With duplication disabled and modify_params False,
+    # we'd expect an empty contents field
+    # This might actually raise an exception in practice
+    assert "system_instruction" in result_no_duplicate
+    
+    # Case 3: With litellm.modify_params=True it should duplicate even with parameter set to False
+    litellm.modify_params = True
+    
+    result_with_modify_params = _transform_request_body(
+        messages=messages.copy(), # type: ignore
+        model=model,
+        optional_params={"duplicate_system_as_user_for_gemini": False},
+        custom_llm_provider=custom_llm_provider,
+        litellm_params={},
+        cached_content=None
+    )
+    
+    # Restore original modify_params value
+    litellm.modify_params = original_modify_params
+    
+    # Verify that contents has user message due to modify_params=True
+    assert len(result_with_modify_params["contents"]) > 0
+    assert result_with_modify_params["contents"][0]["role"] == "user" # type: ignore