diff --git a/litellm/litellm_core_utils/realtime_streaming.py b/litellm/litellm_core_utils/realtime_streaming.py
index 2f2a9ca334a..ae2fae8bd9a 100644
--- a/litellm/litellm_core_utils/realtime_streaming.py
+++ b/litellm/litellm_core_utils/realtime_streaming.py
@@ -525,15 +525,20 @@ class RealTimeStreaming:
             else [transformed_response]
         )
         for event in events:
-            event_str = json.dumps(event)
             if isinstance(event, dict) and event.get("type") == "session.created":
                 if self._session_created_sent_to_client:
+                    # A synthetic session.created (with placeholder defaults) was
+                    # already forwarded to the client when we connected.  The
+                    # provider's real session.created (e.g. emitted from Gemini
+                    # `setupComplete`) carries the authoritative modalities/model
+                    # from the client's session.update.  Re-emit it as
+                    # `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()
-                    verbose_logger.debug(
-                        "Skipping duplicate session.created from provider stream"
-                    )
-                    continue
-                self._session_created_sent_to_client = True
+                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)
diff --git a/litellm/llms/gemini/realtime/transformation.py b/litellm/llms/gemini/realtime/transformation.py
index a987f18a0c7..eb9d4ef0cbd 100644
--- a/litellm/llms/gemini/realtime/transformation.py
+++ b/litellm/llms/gemini/realtime/transformation.py
@@ -1276,6 +1276,12 @@ class GeminiRealtimeConfig(BaseRealtimeConfig):
                     output_items=None,
                 )
                 returned_message.append(transformed_response_done_event)
+                # Reset IDs so a subsequent turn (e.g. a `toolCall` arriving in
+                # a later WebSocket frame after `turnComplete`) starts a fresh
+                # response with its own `response.created` preamble instead of
+                # reusing the just-completed response ID.
+                current_output_item_id = None
+                current_response_id = None
             elif (
                 openai_event == OpenAIRealtimeEventTypes.RESPONSE_TEXT_DELTA
                 or openai_event == OpenAIRealtimeEventTypes.RESPONSE_TEXT_DONE