From 02bb3f9310e6633caefd8f739d2340ebd0d56cff Mon Sep 17 00:00:00 2001
From: Timik232 <100406268+Timik232@users.noreply.github.com>
Date: Mon, 24 Aug 2026 20:56:54 +0300
Subject: [PATCH] fix(streaming): keep response id stable across streamed
 chunks

Providers that stream via GenericStreamingChunk (e.g. GigaChat) do not
propagate an upstream response id, so every chunk of one streamed
response got a freshly generated id. Pin CustomStreamWrapper.response_id
from the first chunk it creates, mirroring the existing 'created'
pinning (#11437).

Clients that merge deltas by chunk id (e.g. goose) split one reply into
one message per chunk.

Fixes #38098
---
 .../litellm_core_utils/streaming_handler.py   |  2 +
 .../test_streaming_handler.py                 | 79 +++++++++++++++++++
 2 files changed, 81 insertions(+)

diff --git a/litellm/litellm_core_utils/streaming_handler.py b/litellm/litellm_core_utils/streaming_handler.py
index f6340426c1b..fb693943b2b 100644
--- a/litellm/litellm_core_utils/streaming_handler.py
+++ b/litellm/litellm_core_utils/streaming_handler.py
@@ -816,6 +816,8 @@ class CustomStreamWrapper:
         model_response: Final = ModelResponseStream(**args)
         if self.response_id is not None:
             model_response.id = self.response_id
+        elif model_response.id:
+            self.response_id = model_response.id
         if self.system_fingerprint is not None:
             model_response.system_fingerprint = self.system_fingerprint
 
diff --git a/tests/test_litellm/litellm_core_utils/test_streaming_handler.py b/tests/test_litellm/litellm_core_utils/test_streaming_handler.py
index b5e33a4e421..a76d427495c 100644
--- a/tests/test_litellm/litellm_core_utils/test_streaming_handler.py
+++ b/tests/test_litellm/litellm_core_utils/test_streaming_handler.py
@@ -4460,3 +4460,82 @@ def test_handle_stream_fallback_error_restores_context_only_after_exception_mapp
     finally:
         trace_id_var.set("")
         session_id_var.set("")
+
+
+class TestStableStreamingResponseId:
+    """
+    All chunks of one streamed response must share the same top-level id
+    (OpenAI streaming contract). Providers streaming via GenericStreamingChunk
+    (e.g. GigaChat) do not propagate an upstream response id, so
+    CustomStreamWrapper must pin the id from the first chunk it creates,
+    mirroring the existing `created` pinning (issue #11437).
+
+    Clients such as goose merge streamed deltas into one assistant message by
+    chunk id; per-chunk ids split a single reply into many messages.
+    """
+
+    def test_generic_chunks_share_one_id(self):
+        def _generic_chunks():
+            return iter(
+                [
+                    {
+                        "text": "Hello",
+                        "tool_use": None,
+                        "is_finished": False,
+                        "finish_reason": "",
+                        "usage": None,
+                        "index": 0,
+                    },
+                    {
+                        "text": " world",
+                        "tool_use": None,
+                        "is_finished": False,
+                        "finish_reason": "",
+                        "usage": None,
+                        "index": 0,
+                    },
+                    {
+                        "text": "",
+                        "tool_use": None,
+                        "is_finished": True,
+                        "finish_reason": "stop",
+                        "usage": {
+                            "prompt_tokens": 1,
+                            "completion_tokens": 2,
+                            "total_tokens": 3,
+                        },
+                        "index": 0,
+                    },
+                ]
+            )
+
+        wrapper = CustomStreamWrapper(
+            completion_stream=_generic_chunks(),
+            model="gigachat/GigaChat-2-Max",
+            logging_obj=MagicMock(),
+            custom_llm_provider="gigachat",
+        )
+        ids = [chunk.id for chunk in wrapper if chunk.id]
+        assert ids, "no chunks emitted"
+        assert len(set(ids)) == 1, f"chunk ids differ across one stream: {ids}"
+
+    def test_creator_pins_id_from_first_chunk(self):
+        wrapper = CustomStreamWrapper(
+            completion_stream=iter([]),
+            model="gigachat/GigaChat-2-Max",
+            logging_obj=MagicMock(),
+            custom_llm_provider="gigachat",
+        )
+        first = wrapper.model_response_creator()
+        assert wrapper.response_id == first.id
+        assert wrapper.model_response_creator().id == first.id
+
+    def test_provider_supplied_id_still_wins(self):
+        wrapper = CustomStreamWrapper(
+            completion_stream=iter([]),
+            model="gigachat/GigaChat-2-Max",
+            logging_obj=MagicMock(),
+            custom_llm_provider="gigachat",
+        )
+        wrapper.response_id = "chatcmpl-from-provider"
+        assert wrapper.model_response_creator().id == "chatcmpl-from-provider"