From 59cae5eb00ea30f321da58c3cefe3de78909d228 Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Fri, 17 Jul 2026 14:08:57 +0000
Subject: [PATCH] fix(logging): preserve prompt_tokens/completion_tokens for
 non-streaming /v1/responses usage

---
 litellm/types/llms/openai.py                  |  3 ++-
 .../test_litellm_logging.py                   | 27 +++++++++++++++++++
 2 files changed, 29 insertions(+), 1 deletion(-)

diff --git a/litellm/types/llms/openai.py b/litellm/types/llms/openai.py
index daac1e4506f..80a36ee2e8f 100644
--- a/litellm/types/llms/openai.py
+++ b/litellm/types/llms/openai.py
@@ -76,6 +76,7 @@ from pydantic import (
     ConfigDict,
     Discriminator,
     PrivateAttr,
+    SerializeAsAny,
     field_serializer,
     field_validator,
 )
@@ -1270,7 +1271,7 @@ class ResponsesAPIResponse(BaseLiteLLMOpenAIResponseObject):
     status: Optional[str] = None
     text: Optional[Union["ResponseText", Dict[str, Any]]] = None
     truncation: Optional[Literal["auto", "disabled"]] = None
-    usage: Optional[ResponseAPIUsage] = None
+    usage: Optional[SerializeAsAny[ResponseAPIUsage]] = None
     user: Optional[str] = None
     store: Optional[bool] = None
     # Define private attributes using PrivateAttr
diff --git a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
index 5bffda126fe..fc6dece21df 100644
--- a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
+++ b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py
@@ -3609,6 +3609,33 @@ def test_handle_anthropic_messages_response_logging_translates_bare_responses_ap
     assert result.usage.total_tokens == 18  # type: ignore[attr-defined]
 
 
+def test_transform_usage_objects_responses_api_keeps_prompt_and_completion_tokens():
+    """Regression for #33688. Non-streaming /v1/responses logging runs the result
+    through _transform_usage_objects, which swaps the ResponseAPIUsage for a chat
+    Usage carrying prompt_tokens/completion_tokens. Serializing the returned
+    ResponsesAPIResponse must preserve those fields; before the SerializeAsAny fix
+    the ResponseAPIUsage-typed field silently dropped them, leaving only
+    total_tokens for downstream loggers (custom callbacks, Prometheus)."""
+    logging_obj = LitellmLogging(
+        model="gpt-4o",
+        messages=[{"role": "user", "content": "hi"}],
+        stream=False,
+        call_type="aresponses",
+        litellm_call_id="test-call-id",
+        start_time=time.time(),
+        function_id="test-fn",
+    )
+
+    result = logging_obj._transform_usage_objects(
+        result=_responses_api_response_with_text("hello world")
+    )
+
+    serialized_usage = result.model_dump()["usage"]
+    assert serialized_usage["prompt_tokens"] == 11
+    assert serialized_usage["completion_tokens"] == 7
+    assert serialized_usage["total_tokens"] == 18
+
+
 def test_handle_anthropic_messages_response_logging_passes_model_response_through():
     """Anthropic-native path already yields a ModelResponse; it must be returned unchanged."""
     logging_obj = _anthropic_messages_logging_obj()