From 9572b60b80c7fbda8d4434a1745efb5a25aeb6f4 Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Mon, 24 Aug 2026 20:09:03 +0000
Subject: [PATCH] refactor(responses): move streaming header restore behind a
 typed model method

Keeps the logged copy's hidden params limited to headers so cost fields on the bridge path are untouched, and avoids private-attr writes from outside the model.

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 litellm/responses/streaming_iterator.py | 14 +++++++++-----
 litellm/types/llms/openai.py            | 10 ++++++++++
 2 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/litellm/responses/streaming_iterator.py b/litellm/responses/streaming_iterator.py
index 127a5301db0..04410827438 100644
--- a/litellm/responses/streaming_iterator.py
+++ b/litellm/responses/streaming_iterator.py
@@ -207,10 +207,11 @@ class BaseResponsesAPIStreamingIterator:
             "api_base": _api_base,
             "custom_llm_provider": custom_llm_provider,
         }
-        self._hidden_params["additional_headers"] = process_response_headers(
-            self.response.headers or {}
-        )  # GUARANTEE OPENAI HEADERS IN RESPONSE
-        self._hidden_params["headers"] = dict(self.response.headers or {})  # mutable-ok: raw headers for logging
+        _raw_headers: Final = dict(self.response.headers or {})  # mutable-ok: process_response_headers takes a dict
+        _processed_headers: Final = process_response_headers(_raw_headers)
+        self._hidden_params["additional_headers"] = _processed_headers  # GUARANTEE OPENAI HEADERS IN RESPONSE
+        self._raw_headers: Final[Mapping[str, str]] = MappingProxyType(_raw_headers)
+        self._processed_headers: Final[Mapping[str, object]] = MappingProxyType(_processed_headers)
 
     def _check_max_streaming_duration(self) -> None:
         """Raise litellm.Timeout if the stream has exceeded LITELLM_MAX_STREAMING_DURATION_SECONDS."""
@@ -425,7 +426,10 @@ class BaseResponsesAPIStreamingIterator:
         copied: Final = self._copy_for_logging(completed)
         inner_response: Final = getattr(copied, "response", None)
         if isinstance(inner_response, ResponsesAPIResponse):
-            inner_response._hidden_params = dict(self._hidden_params)  # mutable-ok: model_dump drops private attrs
+            inner_response.store_provider_response_headers(
+                processed_headers=self._processed_headers,
+                raw_headers=self._raw_headers,
+            )
         return copied
 
     @staticmethod
diff --git a/litellm/types/llms/openai.py b/litellm/types/llms/openai.py
index e7a3f825455..e33e994ad9d 100644
--- a/litellm/types/llms/openai.py
+++ b/litellm/types/llms/openai.py
@@ -1337,6 +1337,16 @@ class ResponsesAPIResponse(BaseLiteLLMOpenAIResponseObject):
     # Define private attributes using PrivateAttr
     _hidden_params: dict = PrivateAttr(default_factory=dict)
 
+    def store_provider_response_headers(
+        self, *, processed_headers: Mapping[str, object], raw_headers: Mapping[str, str]
+    ) -> None:
+        """Keep provider headers on a response rebuilt from `model_dump()`, which drops private attrs."""
+        self._hidden_params = {  # mutable-ok: the private attr is typed as dict
+            **self._hidden_params,
+            "additional_headers": dict(processed_headers),  # mutable-ok: logging callbacks expect a plain dict
+            "headers": dict(raw_headers),  # mutable-ok: logging callbacks expect a plain dict
+        }
+
     @field_validator("reasoning", mode="before")
     @classmethod
     def validate_reasoning_to_dict(cls, value: Any) -> dict[str, Any] | None: