From 13b77ea1b9a92de396538aa13796fe3aee22d84a Mon Sep 17 00:00:00 2001
From: daleselaji-dev <265319989+daleselaji-dev@users.noreply.github.com>
Date: Wed, 26 Aug 2026 10:40:02 +0800
Subject: [PATCH] fix(openai): preserve streaming service tier

---
 .../llms/openai/chat/gpt_transformation.py    |  2 ++
 .../chat/test_openai_gpt_transformation.py    | 19 ++++++++++++++++++-
 2 files changed, 20 insertions(+), 1 deletion(-)

diff --git a/litellm/llms/openai/chat/gpt_transformation.py b/litellm/llms/openai/chat/gpt_transformation.py
index 16fd042cb2f..fc2f9324906 100644
--- a/litellm/llms/openai/chat/gpt_transformation.py
+++ b/litellm/llms/openai/chat/gpt_transformation.py
@@ -807,6 +807,8 @@ class OpenAIChatCompletionStreamingHandler(BaseModelResponseIterator):
             }
             if "usage" in chunk and chunk["usage"] is not None:
                 kwargs["usage"] = chunk["usage"]
+            if "service_tier" in chunk and chunk["service_tier"] is not None:
+                kwargs["service_tier"] = chunk["service_tier"]
             return ModelResponseStream(**kwargs)
         except Exception as e:
             raise e
diff --git a/tests/test_litellm/llms/openai/chat/test_openai_gpt_transformation.py b/tests/test_litellm/llms/openai/chat/test_openai_gpt_transformation.py
index f4c38f8f797..4201e0a9713 100644
--- a/tests/test_litellm/llms/openai/chat/test_openai_gpt_transformation.py
+++ b/tests/test_litellm/llms/openai/chat/test_openai_gpt_transformation.py
@@ -5,7 +5,6 @@ Tests for OpenAI GPT transformation (litellm/llms/openai/chat/gpt_transformation
 
 import pytest
 
-
 import litellm
 from litellm.litellm_core_utils.prompt_templates.common_utils import TOOL_RESULT_IMAGE_BOUNDARY
 from litellm.llms.openai.chat.gpt_5_transformation import OpenAIGPT5Config
@@ -184,6 +183,24 @@ class TestOpenAIChatCompletionStreamingHandler:
         assert result.usage.completion_tokens == 350
         assert result.usage.total_tokens == 14147
 
+    def test_chunk_parser_preserves_service_tier(self):
+        """Provider service-tier responses must survive chunk normalization."""
+        handler = OpenAIChatCompletionStreamingHandler(
+            streaming_response=None, sync_stream=True
+        )
+
+        result = handler.chunk_parser(
+            {
+                "id": "gen-123",
+                "created": 1234567890,
+                "model": "xai/grok-4.6",
+                "choices": [],
+                "service_tier": "priority",
+            }
+        )
+
+        assert result.service_tier == "priority"
+
     def test_chunk_parser_raises_on_in_body_error_payload(self):
         """vLLM/sglang return HTTP 200 streams whose body carries the error,
         e.g. data: {"error": {..., "code": 400}}. chunk_parser must surface it