feat(prometheus): add requested_model label to spend and requests metrics (#31410)

litellm_spend_metric_total and litellm_requests_metric_total previously
exposed only the resolved backend model_id and friendly model name, so
operators could not group spend or request counts by the model alias the
caller actually asked for when a router fronts multiple deployments
behind one name.

This adds the existing UserAPIKeyLabelNames.REQUESTED_MODEL to both
labelname lists; the value is already populated upstream from
standard_logging_payload["model_group"] and flows through the shared
_increment_top_level_request_and_spend_metrics call site. The sibling
token metrics (input/output/total) already carry the label, so this
also restores cross-metric consistency.

Resolves LIT-3796

litellm/types/integrations/prometheus.py

@@ -409,6 +409,7 @@ class PrometheusMetricLabels:
         UserAPIKeyLabelNames.USER_EMAIL.value,
         UserAPIKeyLabelNames.CLIENT_IP.value,
         UserAPIKeyLabelNames.USER_AGENT.value,
+        UserAPIKeyLabelNames.REQUESTED_MODEL.value,
         UserAPIKeyLabelNames.MODEL_ID.value,
         UserAPIKeyLabelNames.API_PROVIDER.value,
     ]
@@ -424,6 +425,7 @@ class PrometheusMetricLabels:
         UserAPIKeyLabelNames.USER_EMAIL.value,
         UserAPIKeyLabelNames.CLIENT_IP.value,
         UserAPIKeyLabelNames.USER_AGENT.value,
+        UserAPIKeyLabelNames.REQUESTED_MODEL.value,
         UserAPIKeyLabelNames.MODEL_ID.value,
         UserAPIKeyLabelNames.API_PROVIDER.value,
     ]

tests/enterprise/litellm_enterprise/enterprise_callbacks/test_prometheus_logging_callbacks.py

@@ -607,6 +607,7 @@ def test_increment_top_level_request_and_spend_metrics(prometheus_logger):
         api_provider="openai",
         client_ip=None,
         user_agent=None,
+        requested_model=None,
     )
     prometheus_logger.litellm_requests_metric.labels().inc.assert_called_once()
 
@@ -626,6 +627,7 @@ def test_increment_top_level_request_and_spend_metrics(prometheus_logger):
         api_provider="openai",
         client_ip=None,
         user_agent=None,
+        requested_model=None,
     )
     prometheus_logger.litellm_spend_metric.labels().inc.assert_called_once_with(0.1)
 

tests/test_litellm/integrations/test_prometheus_labels.py

@@ -148,6 +148,34 @@ def test_model_id_in_required_metrics():
         print(f"✅ {metric_name} contains model_id label")
 
 
+def test_requested_model_in_spend_and_requests_metrics():
+    """
+    Regression test for LIT-3796.
+
+    litellm_spend_metric and litellm_requests_metric must expose the
+    requested_model label so spend and request counts can be grouped by the
+    model alias the caller asked for, not just the backend deployment that
+    served the request. The sibling token metrics (input/output/total)
+    already carry this label; spend and requests were the odd ones out, even
+    though both are emitted side-by-side from the same call site.
+    """
+    requested_model_label = UserAPIKeyLabelNames.REQUESTED_MODEL.value
+
+    metrics_with_requested_model = [
+        "litellm_spend_metric",
+        "litellm_requests_metric",
+        "litellm_input_tokens_metric",
+        "litellm_output_tokens_metric",
+        "litellm_total_tokens_metric",
+    ]
+
+    for metric_name in metrics_with_requested_model:
+        labels = PrometheusMetricLabels.get_labels(metric_name)
+        assert requested_model_label in labels, (
+            f"Metric {metric_name} should contain requested_model label"
+        )
+
+
 def test_route_normalization_for_responses_api():
     """
     Test that route normalization prevents high cardinality in Prometheus metrics