feat(proxy): add x-litellm-model-name response header with deployment model string (#33698)

The proxy already returns x-litellm-model-id (the deployment id) and x-litellm-model-group (the requested model-group alias), but never surfaces the concrete underlying model that served the request; the router rewrites the response model field to the group alias, so callers had no way to read the actual deployment model like anthropic/claude-haiku-4-5. Expose it as x-litellm-model-name, sourced from the deployment recorded in litellm_params metadata.

Co-authored-by: Krrish Dholakia <krrishdholakia@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/proxy/common_request_processing.py

@@ -925,9 +925,12 @@ class ProxyBaseLLMRequestProcessing:
                 # If conversion fails, use original spend
                 pass
 
+        model_name = ProxyBaseLLMRequestProcessing._get_deployment_model_name(litellm_logging_obj)
+
         headers = {
             "x-litellm-call-id": call_id,
             "x-litellm-model-id": model_id,
+            "x-litellm-model-name": model_name,
             "x-litellm-cache-key": cache_key,
             "x-litellm-model-api-base": (
                 api_base.split("?")[0] if api_base else None
@@ -1396,6 +1399,27 @@ class ProxyBaseLLMRequestProcessing:
             model_id = model_info.get("id", "") or ""
         return model_id
 
+    @staticmethod
+    def _get_deployment_model_name(
+        litellm_logging_obj: LiteLLMLoggingObj | None,
+    ) -> str | None:
+        """Extract the underlying deployment model string (e.g. ``azure/gpt-4o``).
+
+        The router rewrites the response ``model`` field to the model-group alias
+        the client requested, so neither the response body nor the existing
+        headers expose the concrete deployment model. The router records it under
+        ``litellm_params`` metadata as ``deployment``, so read it back from there.
+        """
+        litellm_params = getattr(litellm_logging_obj, "litellm_params", None)
+        if not isinstance(litellm_params, dict):
+            return None
+        for key in ("litellm_metadata", "metadata"):
+            metadata = litellm_params.get(key, {}) or {}
+            deployment = metadata.get("deployment")
+            if deployment:
+                return deployment
+        return None
+
     @staticmethod
     def _response_cost_from_logging_obj(
         *,

tests/test_litellm/proxy/test_model_id_header_propagation.py

@@ -200,6 +200,60 @@ def test_get_custom_headers_without_model_id():
         assert headers["x-litellm-model-id"] in [None, ""]
 
 
+class _FakeLoggingObj:
+    def __init__(self, litellm_params):
+        self.litellm_params = litellm_params
+        self.litellm_call_id = "test-call-id"
+
+
+@pytest.mark.parametrize("metadata_key", ["metadata", "litellm_metadata"])
+def test_get_custom_headers_includes_deployment_model_name(metadata_key):
+    """
+    x-litellm-model-name should expose the underlying deployment model string,
+    which the router records under litellm_params[metadata]["deployment"].
+    """
+    mock_user_api_key_dict = MagicMock()
+    mock_user_api_key_dict.tpm_limit = 1000
+    mock_user_api_key_dict.rpm_limit = 100
+
+    logging_obj = _FakeLoggingObj(
+        litellm_params={metadata_key: {"deployment": "azure/gpt-4o-2024-08-06"}}
+    )
+
+    headers = ProxyBaseLLMRequestProcessing.get_custom_headers(
+        user_api_key_dict=mock_user_api_key_dict,
+        model_id="deployment-uuid",
+        request_data={},
+        hidden_params={},
+        litellm_logging_obj=logging_obj,
+    )
+
+    assert headers["x-litellm-model-name"] == "azure/gpt-4o-2024-08-06"
+    assert headers["x-litellm-model-id"] == "deployment-uuid"
+
+
+def test_get_custom_headers_omits_model_name_when_deployment_missing():
+    """
+    Without a deployment model string, x-litellm-model-name must not be emitted
+    (rather than leaking an empty/None value).
+    """
+    mock_user_api_key_dict = MagicMock()
+    mock_user_api_key_dict.tpm_limit = 1000
+    mock_user_api_key_dict.rpm_limit = 100
+
+    logging_obj = _FakeLoggingObj(litellm_params={"metadata": {}})
+
+    headers = ProxyBaseLLMRequestProcessing.get_custom_headers(
+        user_api_key_dict=mock_user_api_key_dict,
+        model_id="deployment-uuid",
+        request_data={},
+        hidden_params={},
+        litellm_logging_obj=logging_obj,
+    )
+
+    assert "x-litellm-model-name" not in headers
+
+
 def test_get_custom_headers_with_empty_string_model_id():
     """
     Test that get_custom_headers handles empty string model_id correctly.