fix(cost): coerce None/string token counts in cost_per_token

litellm/cost_calculator.py

@@ -310,6 +310,17 @@ def _transcription_usage_has_token_details(
     return (prompt_tokens_val > 0) or (completion_tokens_val > 0)
 
 
+def _coerce_token_count(value: float | str | None) -> int:
+    if value is None:
+        return 0
+    if isinstance(value, str):
+        try:
+            return int(value)
+        except ValueError:
+            return 0
+    return int(value)
+
+
 def cost_per_token(
     model: str = "",
     prompt_tokens: int = 0,
@@ -367,6 +378,9 @@ def cost_per_token(
     if model is None:
         raise Exception("Invalid arg. Model cannot be none.")
 
+    prompt_tokens = _coerce_token_count(prompt_tokens)
+    completion_tokens = _coerce_token_count(completion_tokens)
+
     ## RECONSTRUCT USAGE BLOCK ##
     if usage_object is not None:
         usage_block = usage_object

tests/test_litellm/test_cost_calculator.py

@@ -4768,3 +4768,20 @@ def test_collect_and_combine_realtime_usage_stores_partitioned_text_tokens() ->
     assert combined.completion_tokens_details.reasoning_tokens == 95
     assert combined.completion_tokens_details.text_tokens == 38
     assert combined.completion_tokens_details.audio_tokens == 0
+def test_cost_per_token_none_token_counts_cost_zero():
+    prompt_cost, completion_cost = cost_per_token(
+        model="gpt-4o-mini", prompt_tokens=None, completion_tokens=None
+    )
+    assert prompt_cost == 0.0
+    assert completion_cost == 0.0
+
+
+def test_cost_per_token_string_token_counts_coerced():
+    prompt_cost, completion_cost = cost_per_token(
+        model="gpt-4o-mini", prompt_tokens="10", completion_tokens=5
+    )
+    expected_prompt, expected_completion = cost_per_token(
+        model="gpt-4o-mini", prompt_tokens=10, completion_tokens=5
+    )
+    assert prompt_cost == expected_prompt
+    assert completion_cost == expected_completion