diff --git a/litellm/tests/test_utils.py b/litellm/tests/test_utils.py
index 976ded7f60..6b0148b4e0 100644
--- a/litellm/tests/test_utils.py
+++ b/litellm/tests/test_utils.py
@@ -761,3 +761,16 @@ def test_supports_response_schema(model, expected_bool):
     response = supports_response_schema(model=model, custom_llm_provider=None)
 
     assert expected_bool == response
+
+
+def test_usage_object_null_tokens():
+    """
+    Unit test.
+
+    Asserts Usage obj always returns int.
+
+    Fixes https://github.com/BerriAI/litellm/issues/5096
+    """
+    usage_obj = litellm.Usage(prompt_tokens=2, completion_tokens=None, total_tokens=2)
+
+    assert usage_obj.completion_tokens == 0
diff --git a/litellm/types/utils.py b/litellm/types/utils.py
index 1a05c1d2a0..361163e6a5 100644
--- a/litellm/types/utils.py
+++ b/litellm/types/utils.py
@@ -444,11 +444,10 @@ class Choices(OpenAIObject):
         setattr(self, key, value)
 
 
-class Usage(OpenAIObject):
-    prompt_tokens: Optional[int] = Field(default=None)
-    completion_tokens: Optional[int] = Field(default=None)
-    total_tokens: Optional[int] = Field(default=None)
+from openai.types.completion_usage import CompletionUsage
 
+
+class Usage(CompletionUsage):
     def __init__(
         self,
         prompt_tokens: Optional[int] = None,
@@ -457,14 +456,16 @@ class Usage(OpenAIObject):
         **params,
     ):
         data = {
-            "prompt_tokens": prompt_tokens,
-            "completion_tokens": completion_tokens,
-            "total_tokens": total_tokens,
-            **params,
+            "prompt_tokens": prompt_tokens or 0,
+            "completion_tokens": completion_tokens or 0,
+            "total_tokens": total_tokens or 0,
         }
 
         super().__init__(**data)
 
+        for k, v in params.items():
+            setattr(self, k, v)
+
     def __contains__(self, key):
         # Define custom behavior for the 'in' operator
         return hasattr(self, key)