diff --git a/litellm/proxy/common_request_processing.py b/litellm/proxy/common_request_processing.py
index dc8bc961296..37cc3fab720 100644
--- a/litellm/proxy/common_request_processing.py
+++ b/litellm/proxy/common_request_processing.py
@@ -2605,6 +2605,13 @@ class ProxyBaseLLMRequestProcessing:
     ):
         """Raises ProxyException (OpenAI API compatible) if an exception is raised"""
         _log_llm_api_exception(e)
+        _litellm_logging_obj: Optional[LiteLLMLoggingObj] = self.data.get("litellm_logging_obj", None)
+
+        # Attempt to get model_id from logging object
+        #
+        # Note: We check the direct model_info path first (not nested in metadata) because that's where the router sets it.
+        # The nested metadata path is only a fallback for cases where model_info wasn't set at the top level.
+        model_id = self.maybe_get_model_id(_litellm_logging_obj)
         # Allow callbacks to transform the error response
         transformed_exception = await proxy_logging_obj.post_call_failure_hook(
             user_api_key_dict=user_api_key_dict,
@@ -2624,13 +2631,6 @@ class ProxyBaseLLMRequestProcessing:
         timeout = getattr(
             e, "timeout", None
         )  # returns the timeout set by the wrapper. Used for testing if model-specific timeout are set correctly
-        _litellm_logging_obj: Optional[LiteLLMLoggingObj] = self.data.get("litellm_logging_obj", None)
-
-        # Attempt to get model_id from logging object
-        #
-        # Note: We check the direct model_info path first (not nested in metadata) because that's where the router sets it.
-        # The nested metadata path is only a fallback for cases where model_info wasn't set at the top level.
-        model_id = self.maybe_get_model_id(_litellm_logging_obj)
 
         custom_headers = ProxyBaseLLMRequestProcessing.get_custom_headers(
             user_api_key_dict=user_api_key_dict,
diff --git a/tests/test_litellm/proxy/test_common_request_processing.py b/tests/test_litellm/proxy/test_common_request_processing.py
index 3bb84e095a0..8a59478be31 100644
--- a/tests/test_litellm/proxy/test_common_request_processing.py
+++ b/tests/test_litellm/proxy/test_common_request_processing.py
@@ -2948,6 +2948,47 @@ class TestHandleLLMApiExceptionRetryAfter:
         assert proxy_exc.headers["x-custom"] == "1"
 
 
+class TestHandleLLMApiExceptionModelIdHeader:
+    """post_call_failure_hook pops litellm_logging_obj off request_data, so the
+    error handler must resolve model_id before invoking it or x-litellm-model-id
+    is dropped from error responses (the reported embeddings 429 bug)."""
+
+    class _FakeLoggingObj:
+        def __init__(self, model_id: str):
+            self.litellm_params = {"model_info": {"id": model_id}}
+            self.litellm_call_id = "call-123"
+            self.kwargs: dict = {}
+
+    async def test_model_id_survives_failure_hook_that_pops_logging_obj(self):
+        from litellm.proxy._types import ProxyException, UserAPIKeyAuth
+
+        data = {
+            "model": "text-embedding-ada-002",
+            "litellm_call_id": "call-123",
+            "litellm_logging_obj": self._FakeLoggingObj("deployment-xyz"),
+        }
+        processor = ProxyBaseLLMRequestProcessing(data=data)
+
+        async def _popping_failure_hook(request_data, **kwargs):
+            request_data.pop("litellm_logging_obj", None)
+            return None
+
+        proxy_logging_obj = MagicMock()
+        proxy_logging_obj.post_call_failure_hook = AsyncMock(
+            side_effect=_popping_failure_hook
+        )
+        proxy_logging_obj.post_call_response_headers_hook = AsyncMock(return_value={})
+
+        with pytest.raises(ProxyException) as exc_info:
+            await processor._handle_llm_api_exception(
+                e=ValueError("429 rate limit"),
+                user_api_key_dict=UserAPIKeyAuth(api_key="sk-test"),
+                proxy_logging_obj=proxy_logging_obj,
+            )
+
+        assert exc_info.value.headers["x-litellm-model-id"] == "deployment-xyz"
+
+
 class TestAsyncStreamingDataGeneratorFastPath:
     """Fast/slow path branching in async_streaming_data_generator."""