diff --git a/litellm/integrations/prometheus.py b/litellm/integrations/prometheus.py
index f9b1c666439..d519d68d346 100644
--- a/litellm/integrations/prometheus.py
+++ b/litellm/integrations/prometheus.py
@@ -1444,9 +1444,23 @@ class PrometheusLogger(CustomLogger):
         remaining_tokens_variable_name = f"litellm-key-remaining-tokens-{model_group}"
 
         remaining_requests = metadata.get(remaining_requests_variable_name)
+        if remaining_requests is None:
+            remaining_requests = (
+                self._get_model_per_key_remaining_value_from_additional_headers(
+                    kwargs=kwargs,
+                    rate_limit_type="requests",
+                )
+            )
         if remaining_requests is None:
             remaining_requests = sys.maxsize
         remaining_tokens = metadata.get(remaining_tokens_variable_name)
+        if remaining_tokens is None:
+            remaining_tokens = (
+                self._get_model_per_key_remaining_value_from_additional_headers(
+                    kwargs=kwargs,
+                    rate_limit_type="tokens",
+                )
+            )
         if remaining_tokens is None:
             remaining_tokens = sys.maxsize
 
@@ -1484,6 +1498,28 @@ class PrometheusLogger(CustomLogger):
             remaining_tokens
         )
 
+    @staticmethod
+    def _get_model_per_key_remaining_value_from_additional_headers(
+        kwargs: dict,
+        rate_limit_type: Literal["requests", "tokens"],
+    ) -> Optional[Any]:
+        standard_logging_payload = kwargs.get("standard_logging_object") or {}
+        hidden_params = standard_logging_payload.get("hidden_params") or {}
+        additional_headers = hidden_params.get("additional_headers") or {}
+        if not isinstance(additional_headers, dict):
+            return None
+
+        header_name = f"x-ratelimit-model_per_key-remaining-{rate_limit_type}"
+        if header_name in additional_headers:
+            return additional_headers[header_name]
+
+        normalized_header_name = header_name.lower()
+        for key, value in additional_headers.items():
+            if str(key).lower() == normalized_header_name:
+                return value
+
+        return None
+
     def _set_latency_metrics(
         self,
         kwargs: dict,
diff --git a/tests/test_litellm/integrations/test_prometheus_custom_metadata_label_counts.py b/tests/test_litellm/integrations/test_prometheus_custom_metadata_label_counts.py
index 99eb5abb7b5..8dcac51d506 100644
--- a/tests/test_litellm/integrations/test_prometheus_custom_metadata_label_counts.py
+++ b/tests/test_litellm/integrations/test_prometheus_custom_metadata_label_counts.py
@@ -157,3 +157,81 @@ def test_virtual_key_rate_limit_metrics_preserve_zero_remaining_values(
     assert any(sample.value == 0 for sample in token_samples)
     assert not any(sample.value == sys.maxsize for sample in request_samples)
     assert not any(sample.value == sys.maxsize for sample in token_samples)
+
+
+def test_virtual_key_rate_limit_metrics_read_remaining_values_from_hidden_headers(
+    monkeypatch: pytest.MonkeyPatch,
+):
+    prometheus_logger = _create_prometheus_logger_with_custom_labels(monkeypatch)
+    metadata = {
+        "model_group": "gpt-4o-mini",
+    }
+    kwargs = {
+        "litellm_params": {
+            "metadata": metadata,
+        },
+        "standard_logging_object": {
+            **_standard_logging_payload_with_requester_metadata(),
+            "hidden_params": {
+                "additional_headers": {
+                    "x-ratelimit-model_per_key-remaining-requests": 7,
+                    "x-ratelimit-model_per_key-remaining-tokens": 321,
+                },
+            },
+        },
+    }
+
+    prometheus_logger._set_virtual_key_rate_limit_metrics(
+        user_api_key="test-hash",
+        user_api_key_alias="test-alias",
+        kwargs=kwargs,
+        metadata=metadata,
+        model_id="model-123",
+    )
+
+    request_samples = _metric_samples("litellm_remaining_api_key_requests_for_model")
+    token_samples = _metric_samples("litellm_remaining_api_key_tokens_for_model")
+
+    assert any(sample.value == 7 for sample in request_samples)
+    assert any(sample.value == 321 for sample in token_samples)
+    assert not any(sample.value == sys.maxsize for sample in request_samples)
+    assert not any(sample.value == sys.maxsize for sample in token_samples)
+
+
+def test_virtual_key_rate_limit_metrics_preserve_zero_values_from_hidden_headers(
+    monkeypatch: pytest.MonkeyPatch,
+):
+    prometheus_logger = _create_prometheus_logger_with_custom_labels(monkeypatch)
+    metadata = {
+        "model_group": "gpt-4o-mini",
+    }
+    kwargs = {
+        "litellm_params": {
+            "metadata": metadata,
+        },
+        "standard_logging_object": {
+            **_standard_logging_payload_with_requester_metadata(),
+            "hidden_params": {
+                "additional_headers": {
+                    "x-ratelimit-model_per_key-remaining-requests": 0,
+                    "x-ratelimit-model_per_key-remaining-tokens": 0,
+                },
+            },
+        },
+    }
+
+    prometheus_logger._set_virtual_key_rate_limit_metrics(
+        user_api_key="test-hash",
+        user_api_key_alias="test-alias",
+        kwargs=kwargs,
+        metadata=metadata,
+        model_id="model-123",
+    )
+
+    request_samples = _metric_samples("litellm_remaining_api_key_requests_for_model")
+    token_samples = _metric_samples("litellm_remaining_api_key_tokens_for_model")
+
+    assert any(sample.value == 0 for sample in request_samples)
+    assert any(sample.value == 0 for sample in token_samples)
+    assert not any(sample.value == sys.maxsize for sample in request_samples)
+    assert not any(sample.value == sys.maxsize for sample in token_samples)