From b9bf5d82db6720144b1f69b6b8a343b2588411a9 Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Sun, 9 Aug 2026 01:48:25 +0000
Subject: [PATCH] refactor(responses): build split chunks without mutating
 copies

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 .../streaming_iterator.py                     | 19 +++++++++++--------
 .../test_reasoning_content_transformation.py  | 12 ++++++++----
 2 files changed, 19 insertions(+), 12 deletions(-)

diff --git a/litellm/responses/litellm_completion_transformation/streaming_iterator.py b/litellm/responses/litellm_completion_transformation/streaming_iterator.py
index f88883586ab..8ff8293f21b 100644
--- a/litellm/responses/litellm_completion_transformation/streaming_iterator.py
+++ b/litellm/responses/litellm_completion_transformation/streaming_iterator.py
@@ -150,14 +150,17 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
         if not (getattr(delta, "reasoning_content", None) and delta.content):
             return (chunk,)
 
-        reasoning_part: Final = chunk.model_copy(deep=True)
-        reasoning_part.choices[0].delta.content = None
-        reasoning_part.choices[0].finish_reason = None
-
-        content_part: Final = chunk.model_copy(deep=True)
-        content_part.choices[0].delta.reasoning_content = None
-        content_part.choices[0].delta.thinking_blocks = None
-        return (reasoning_part, content_part)
+        choice: Final = chunk.choices[0]
+        reasoning_choice: Final = choice.model_copy(
+            update={"delta": delta.model_copy(update={"content": None}), "finish_reason": None}
+        )
+        content_choice: Final = choice.model_copy(
+            update={"delta": delta.model_copy(update={"reasoning_content": None, "thinking_blocks": None})}
+        )
+        return (
+            chunk.model_copy(update={"choices": [reasoning_choice]}),
+            chunk.model_copy(update={"choices": [content_choice]}),
+        )
 
     def _queue_reasoning_lifecycle_events(self, chunk: ModelResponseStream) -> None:
         if not self._reasoning_active or self._reasoning_done_emitted:
diff --git a/tests/test_litellm/responses/litellm_completion_transformation/test_reasoning_content_transformation.py b/tests/test_litellm/responses/litellm_completion_transformation/test_reasoning_content_transformation.py
index 50a4707555f..14b6f9c3d1f 100644
--- a/tests/test_litellm/responses/litellm_completion_transformation/test_reasoning_content_transformation.py
+++ b/tests/test_litellm/responses/litellm_completion_transformation/test_reasoning_content_transformation.py
@@ -270,7 +270,7 @@ class TestReasoningContentFinalResponse:
 class _FakeChatCompletionStream:
     """Minimal CustomStreamWrapper stand-in that replays pre-built chunks."""
 
-    def __init__(self, chunks):
+    def __init__(self, chunks: list[ModelResponseStream]):
         self._chunks = list(chunks)
         self.logging_obj = SimpleNamespace(
             _response_cost_calculator=lambda result: 0.0,
@@ -294,8 +294,12 @@ class _FakeChatCompletionStream:
         return self._chunks.pop(0)
 
 
-def _reasoning_and_text_chunks():
-    def make(content=None, reasoning=None, finish_reason=None):
+def _reasoning_and_text_chunks() -> list[ModelResponseStream]:
+    def make(
+        content: str | None = None,
+        reasoning: str | None = None,
+        finish_reason: str | None = None,
+    ) -> ModelResponseStream:
         return ModelResponseStream(
             id="chatcmpl-combined",
             created=1234567890,
@@ -320,7 +324,7 @@ def _reasoning_and_text_chunks():
     ]
 
 
-def _new_iterator(chunks):
+def _new_iterator(chunks: list[ModelResponseStream]) -> LiteLLMCompletionStreamingIterator:
     return LiteLLMCompletionStreamingIterator(
         model="test-model",
         litellm_custom_stream_wrapper=_FakeChatCompletionStream(chunks),