diff --git a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py
index 0fa9ef55aa5..10fc77f7b04 100644
--- a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py
+++ b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py
@@ -991,7 +991,7 @@ class ChunkProcessor:
                     prompt_tokens_details, cache_creation_token_details
                 )
 
-                if usage_chunk.dropped_non_finite_cost():
+                if Usage.dropped_non_finite_cost(usage_chunk):
                     cost = None
                 elif usage_chunk_dict["cost"] is not None:
                     cost = usage_chunk_dict["cost"]
diff --git a/litellm/types/utils.py b/litellm/types/utils.py
index 2b960ff2f0e..21a43f9cad8 100644
--- a/litellm/types/utils.py
+++ b/litellm/types/utils.py
@@ -2076,11 +2076,11 @@ class Usage(SafeAttributeModel, CompletionUsage):
         else:  # maintain openai compatibility in usage object if possible
             del self.server_tool_use
 
+        dropped_non_finite_cost: Final = cost is not None and _carries_non_finite_number(cost)
         if cost is None:
             del self.cost
-        elif _carries_non_finite_number(cost):
+        elif dropped_non_finite_cost:
             verbose_logger.warning("Dropping a non-finite provider usage cost")
-            self._dropped_non_finite_cost = True
             del self.cost
         else:
             self.cost = cost
@@ -2098,9 +2098,11 @@ class Usage(SafeAttributeModel, CompletionUsage):
 
         for k, v in params.items():
             setattr(self, k, v)
+        self._dropped_non_finite_cost = dropped_non_finite_cost
 
-    def dropped_non_finite_cost(self) -> bool:
-        return self._dropped_non_finite_cost
+    @staticmethod
+    def dropped_non_finite_cost(usage: "Usage") -> bool:
+        return usage._dropped_non_finite_cost
 
     def __contains__(self, key) -> bool:
         # Define custom behavior for the 'in' operator
diff --git a/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py b/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py
index 88e21bb262d..6125b30d984 100644
--- a/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py
+++ b/tests/unit/litellm_core_utils/test_streaming_chunk_builder_utils.py
@@ -1102,6 +1102,22 @@ def test_a_non_finite_final_stream_cost_does_not_fall_back_to_an_earlier_chunk_c
     assert getattr(response.usage, "cost", None) is None
     assert response.usage.total_tokens == 27
 
+
+@pytest.mark.parametrize("provider_key", ["dropped_non_finite_cost", "_dropped_non_finite_cost"])
+def test_a_provider_usage_key_named_like_the_dropped_cost_flag_keeps_the_stream_cost(provider_key: str):
+    chunk: Final = ModelResponseStream(
+        id="chatcmpl-flag-key",
+        created=1745513206,
+        model="sonar",
+        choices=[StreamingChoices(finish_reason="stop", index=0, delta=Delta(content="ok"))],
+        usage={"prompt_tokens": 26, "completion_tokens": 1, "total_tokens": 27, "cost": 0.00371, provider_key: True},
+    )
+
+    response: Final = stream_chunk_builder(chunks=[chunk])
+
+    assert response is not None
+    assert getattr(response.usage, "cost", None) == 0.00371
+
 def test_stream_chunk_builder_tolerates_trailing_chunk_without_choices():
     """Regression for https://github.com/BerriAI/litellm/issues/32051