From 9678c723b094d10a99ab02e1301404a1095effaa Mon Sep 17 00:00:00 2001
From: Sameer Kankute <sameer@berri.ai>
Date: Wed, 18 Feb 2026 16:47:05 +0530
Subject: [PATCH] Add reasoning' field to 'reasoning_content' field in delta

---
 .../llms/openai/chat/gpt_transformation.py    | 24 +++++-
 .../chat/test_openai_gpt_transformation.py    | 77 ++++++++++++++++++-
 2 files changed, 99 insertions(+), 2 deletions(-)

diff --git a/litellm/llms/openai/chat/gpt_transformation.py b/litellm/llms/openai/chat/gpt_transformation.py
index 5b9840d95b0..59f52e2b81c 100644
--- a/litellm/llms/openai/chat/gpt_transformation.py
+++ b/litellm/llms/openai/chat/gpt_transformation.py
@@ -770,14 +770,36 @@ class OpenAIGPTConfig(BaseLLMModelInfo, BaseConfig):
 
 
 class OpenAIChatCompletionStreamingHandler(BaseModelResponseIterator):
+    def _map_reasoning_to_reasoning_content(self, choices: list) -> list:
+        """
+        Map 'reasoning' field to 'reasoning_content' field in delta.
+        
+        Some OpenAI-compatible providers (e.g., GLM-5, hosted_vllm) return 
+        delta.reasoning, but LiteLLM expects delta.reasoning_content.
+        
+        Args:
+            choices: List of choice objects from the streaming chunk
+            
+        Returns:
+            List of choices with reasoning field mapped to reasoning_content
+        """
+        for choice in choices:
+            delta = choice.get("delta", {})
+            if "reasoning" in delta:
+                delta["reasoning_content"] = delta.pop("reasoning")
+        return choices
+    
     def chunk_parser(self, chunk: dict) -> ModelResponseStream:
         try:
+            choices = chunk.get("choices", [])
+            choices = self._map_reasoning_to_reasoning_content(choices)
+            
             kwargs = {
                 "id": chunk["id"],
                 "object": "chat.completion.chunk",
                 "created": chunk.get("created"),
                 "model": chunk.get("model"),
-                "choices": chunk.get("choices", []),
+                "choices": choices,
             }
             if "usage" in chunk and chunk["usage"] is not None:
                 kwargs["usage"] = chunk["usage"]
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 e6ab199168d..37959f74086 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
@@ -10,8 +10,8 @@ import pytest
 sys.path.insert(0, os.path.abspath("../../../../.."))
 
 from litellm.llms.openai.chat.gpt_transformation import (
-    OpenAIGPTConfig,
     OpenAIChatCompletionStreamingHandler,
+    OpenAIGPTConfig,
 )
 
 
@@ -204,6 +204,81 @@ class TestOpenAIChatCompletionStreamingHandler:
         assert result.choices[0].delta.content == "Hello"
         assert not hasattr(result, "usage") or result.usage is None
 
+    def test_chunk_parser_maps_reasoning_to_reasoning_content(self):
+        """
+        Test that chunk_parser maps 'reasoning' field to 'reasoning_content'.
+        
+        Some OpenAI-compatible providers (e.g., GLM-5, hosted_vllm) return
+        delta.reasoning, but LiteLLM expects delta.reasoning_content.
+        
+        Regression test for: Streaming responses with delta.reasoning field
+        coming back empty when using openai/ or hosted_vllm/ providers.
+        """
+        handler = OpenAIChatCompletionStreamingHandler(
+            streaming_response=None, sync_stream=True
+        )
+
+        # Simulate a chunk with reasoning field (as returned by GLM-5)
+        chunk = {
+            "id": "chatcmpl-8e3d624de9b12528",
+            "object": "chat.completion.chunk",
+            "created": 1771411455,
+            "model": "glm-5",
+            "choices": [
+                {
+                    "index": 0,
+                    "delta": {
+                        "reasoning": "The capital of France",
+                        "role": None,
+                    },
+                    "finish_reason": None,
+                }
+            ],
+        }
+
+        # Parse the chunk
+        parsed_chunk = handler.chunk_parser(chunk)
+
+        # Verify that reasoning was mapped to reasoning_content
+        assert parsed_chunk.choices[0].delta.reasoning_content == "The capital of France"
+        # Verify that the original 'reasoning' field was removed
+        assert not hasattr(parsed_chunk.choices[0].delta, "reasoning")
+
+    def test_chunk_parser_reasoning_field_not_present(self):
+        """
+        Test that chunks without reasoning field still work correctly.
+        """
+        handler = OpenAIChatCompletionStreamingHandler(
+            streaming_response=None, sync_stream=True
+        )
+
+        # Simulate a chunk without reasoning field
+        chunk = {
+            "id": "chatcmpl-test",
+            "object": "chat.completion.chunk",
+            "created": 1769511767,
+            "model": "gpt-4o",
+            "choices": [
+                {
+                    "delta": {
+                        "content": "Regular content",
+                        "role": "assistant",
+                    },
+                    "finish_reason": None,
+                    "index": 0,
+                }
+            ],
+        }
+
+        # Parse the chunk
+        parsed_chunk = handler.chunk_parser(chunk)
+
+        # Verify that content is present
+        assert parsed_chunk.choices[0].delta.content == "Regular content"
+        assert parsed_chunk.choices[0].delta.role == "assistant"
+        # Verify that reasoning_content is not set (it should be deleted by Delta.__init__)
+        assert not hasattr(parsed_chunk.choices[0].delta, "reasoning_content")
+
 
 class TestPromptCacheKeyIntegration:
     """Tests for prompt_cache_key support"""