From dd8b4b1d5460be74056becc8708f8494e67d8ca2 Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Thu, 9 Jul 2026 12:08:47 +0000
Subject: [PATCH] fix(responses-bridge): latch a single chunk id across bridged
 streaming chunks

---
 .../transformation.py                         |  8 ++-
 ...responses_transformation_transformation.py | 72 +++++++++++++++++++
 2 files changed, 79 insertions(+), 1 deletion(-)

diff --git a/litellm/completion_extras/litellm_responses_transformation/transformation.py b/litellm/completion_extras/litellm_responses_transformation/transformation.py
index aecb2552b53..dd646d0d9b1 100644
--- a/litellm/completion_extras/litellm_responses_transformation/transformation.py
+++ b/litellm/completion_extras/litellm_responses_transformation/transformation.py
@@ -1074,6 +1074,7 @@ class LiteLLMResponsesTransformationHandler(CompletionTransformationBridge):
 class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator):
     def __init__(self, streaming_response, sync_stream: bool, json_mode: Optional[bool] = False):
         super().__init__(streaming_response, sync_stream, json_mode)
+        self._chat_completion_id: Optional[str] = None
 
     def _handle_string_chunk(
         self, str_line: Union[str, "BaseModel"]
@@ -1381,4 +1382,9 @@ class OpenAiResponsesToChatCompletionStreamIterator(BaseModelResponseIterator):
             ModelResponseStream: OpenAI-formatted streaming chunk
         """
         verbose_logger.debug(f"Chat provider: transform_streaming_response called with chunk: {chunk}")
-        return OpenAiResponsesToChatCompletionStreamIterator.translate_responses_chunk_to_openai_stream(chunk)
+        parsed_chunk = OpenAiResponsesToChatCompletionStreamIterator.translate_responses_chunk_to_openai_stream(chunk)
+        if self._chat_completion_id is None and parsed_chunk.id:
+            self._chat_completion_id = parsed_chunk.id
+        elif self._chat_completion_id is not None:
+            parsed_chunk.id = self._chat_completion_id
+        return parsed_chunk
diff --git a/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py b/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py
index 6a1de0586dd..8b758ea74be 100644
--- a/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py
+++ b/tests/test_litellm/completion_extras/litellm_responses_transformation/test_completion_extras_litellm_responses_transformation_transformation.py
@@ -2401,6 +2401,78 @@ def test_parallel_tool_calls_comprehensive_streaming_integration():
     )
 
 
+def test_bridge_streaming_chunks_share_single_id():
+    """Regression test for #32607.
+
+    When a chat completions request is bridged to the Responses API and the
+    streaming events are translated back, every emitted chunk must share one
+    stable id. Previously each ModelResponseStream got a fresh
+    "chatcmpl-<uuid>", which broke clients that group tool-call deltas by
+    chunk.id (splitting a single tool call across multiple messages).
+    """
+    from litellm.completion_extras.litellm_responses_transformation.transformation import (
+        OpenAiResponsesToChatCompletionStreamIterator,
+    )
+
+    chunks = [
+        {"type": "response.created", "response": {"id": "resp_001"}},
+        {
+            "type": "response.output_item.added",
+            "output_index": 0,
+            "item": {
+                "type": "function_call",
+                "id": "fc_001",
+                "call_id": "call_1",
+                "name": "get_weather",
+            },
+        },
+        {
+            "type": "response.function_call_arguments.delta",
+            "output_index": 0,
+            "delta": '{"city":"Amsterdam"}',
+        },
+        {
+            "type": "response.output_item.done",
+            "output_index": 0,
+            "item": {
+                "type": "function_call",
+                "id": "fc_001",
+                "call_id": "call_1",
+                "name": "get_weather",
+                "arguments": '{"city":"Amsterdam"}',
+            },
+        },
+        {
+            "type": "response.completed",
+            "response": {
+                "id": "resp_001",
+                "status": "completed",
+                "output": [
+                    {
+                        "type": "function_call",
+                        "call_id": "call_1",
+                        "name": "get_weather",
+                        "arguments": '{"city":"Amsterdam"}',
+                    }
+                ],
+            },
+        },
+    ]
+
+    iterator = OpenAiResponsesToChatCompletionStreamIterator(
+        streaming_response=None, sync_stream=True
+    )
+    ids = [iterator.chunk_parser(chunk).id for chunk in chunks]
+
+    assert len(set(ids)) == 1, (
+        f"All bridged streaming chunks must share one id, got {len(set(ids))} "
+        f"distinct ids: {set(ids)}"
+    )
+    assert ids[0] and ids[0].startswith("chatcmpl-"), (
+        f"Bridged chunk id must keep the chat completion prefix, got {ids[0]!r}"
+    )
+
+
 def test_map_optional_params_preserves_reasoning_summary():
     """Test that reasoning_effort dict with summary field is preserved.