diff --git a/litellm/litellm_core_utils/realtime_streaming.py b/litellm/litellm_core_utils/realtime_streaming.py
index ae2fae8bd9a..a44e6cff1f7 100644
--- a/litellm/litellm_core_utils/realtime_streaming.py
+++ b/litellm/litellm_core_utils/realtime_streaming.py
@@ -525,7 +525,10 @@ class RealTimeStreaming:
             else [transformed_response]
         )
         for event in events:
-            if isinstance(event, dict) and event.get("type") == "session.created":
+            is_session_created_event = (
+                isinstance(event, dict) and event.get("type") == "session.created"
+            )
+            if is_session_created_event:
                 if self._session_created_sent_to_client:
                     # A synthetic session.created (with placeholder defaults) was
                     # already forwarded to the client when we connected.  The
@@ -535,16 +538,14 @@ class RealTimeStreaming:
                     # `session.updated` so the client learns the corrected
                     # configuration without seeing two `session.created` events.
                     event = {**event, "type": "session.updated"}
-                    await self._maybe_send_guardrail_turn_detection_update()
                 else:
                     self._session_created_sent_to_client = True
             event_str = json.dumps(event)
-            ## For audio/VAD guardrail path: forward session.created first, then inject.
-            if (
-                isinstance(event, dict)
-                and event.get("type") == "session.created"
-                and self._has_audio_transcription_guardrails()
-            ):
+            ## For audio/VAD guardrail path: forward the (possibly re-typed)
+            ## session.created first, then inject the guardrail turn-detection
+            ## update. Handling first and duplicate session.created via the same
+            ## path keeps the event-then-guardrail ordering consistent.
+            if is_session_created_event and self._has_audio_transcription_guardrails():
                 self.store_message(event_str)
                 await self.websocket.send_text(event_str)
                 await self._maybe_send_guardrail_turn_detection_update()
diff --git a/litellm/llms/vertex_ai/realtime/transformation.py b/litellm/llms/vertex_ai/realtime/transformation.py
index 555b04ab883..3ef829b24f6 100644
--- a/litellm/llms/vertex_ai/realtime/transformation.py
+++ b/litellm/llms/vertex_ai/realtime/transformation.py
@@ -155,6 +155,21 @@ class VertexAIRealtimeConfig(GeminiRealtimeConfig):
 
     def _build_vertex_ai_setup_config(self, model: str, session_params: dict) -> dict:
         """Build Vertex AI setup configuration with proper model path and defaults."""
+        # Normalize ``turn_detection`` to the top level so map_openai_params
+        # picks it up whether the client used the flat beta shape or the
+        # GA nested shape (session.audio.input.turn_detection). Without
+        # this, guardrail-injected ``create_response: False`` would be
+        # silently dropped for GA clients because map_openai_params only
+        # looks at top-level keys.
+        extracted_turn_detection = self._extract_turn_detection(session_params)
+        if (
+            extracted_turn_detection is not None
+            and "turn_detection" not in session_params
+        ):
+            session_params = {
+                **session_params,
+                "turn_detection": extracted_turn_detection,
+            }
         setup_config = self.map_openai_params(
             optional_params={}, non_default_params=session_params
         )