From 7d5cb8ebb287ea4d9daec840d15b5210dfafa037 Mon Sep 17 00:00:00 2001
From: Cesar Garcia <128240629+Chesars@users.noreply.github.com>
Date: Thu, 20 Nov 2025 00:09:37 -0300
Subject: [PATCH] fix(gemini): Add reasoning_content to streaming responses
 with tools (#16854)

Fixes #16805

When using Gemini models (2.5/3.0) with streaming + tools enabled,
the reasoning_content field was missing from stream chunks, even though
thinking_blocks were present in non-streaming responses.

Changes:
- Convert thinking_blocks to reasoning_content for streaming responses
- Extract "thinking" field from each thinking_block
- Concatenate multiple thinking parts with newlines
- Assign to reasoning_content in chat_completion_message for streaming

Testing:
- Added test_streaming_chunk_with_tool_calls_includes_reasoning_content
- Test verifies reasoning_content appears with tool calls in streaming
- All 39 existing Gemini tests pass
---
 .../vertex_and_google_ai_studio_gemini.py     | 13 +++++
 ...test_vertex_and_google_ai_studio_gemini.py | 52 +++++++++++++++++++
 2 files changed, 65 insertions(+)

diff --git a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
index d83096b26b..72cd8116f9 100644
--- a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
+++ b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
@@ -1673,6 +1673,19 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
             if thinking_blocks is not None:
                 chat_completion_message["thinking_blocks"] = thinking_blocks  # type: ignore
 
+                # Convert thinking_blocks to reasoning_content for streaming
+                # This ensures reasoning_content is available in streaming responses
+                if isinstance(model_response, ModelResponseStream) and reasoning_content is None:
+                    reasoning_content_parts = []
+                    for block in thinking_blocks:
+                        thinking_text = block.get("thinking")
+                        if thinking_text:
+                            reasoning_content_parts.append(thinking_text)
+
+                    if reasoning_content_parts:
+                        reasoning_content = "\n".join(reasoning_content_parts)
+                        chat_completion_message["reasoning_content"] = reasoning_content
+
             if isinstance(model_response, ModelResponseStream):
                 choice = VertexGeminiConfig._create_streaming_choice(
                     chat_completion_message=chat_completion_message,
diff --git a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
index 2e0e9b4d01..5de6bbbced 100644
--- a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
+++ b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
@@ -390,6 +390,58 @@ def test_streaming_chunk_includes_reasoning_content():
     )
 
 
+def test_streaming_chunk_with_tool_calls_includes_reasoning_content():
+    """
+    Test for issue #16805: Ensure that when Gemini returns a streaming chunk with
+    tool calls AND thoughtSignature, the reasoning_content is included in the delta.
+
+    Previously, thinking_blocks were only added to non-streaming responses, causing
+    reasoning_content to be missing in streaming mode when tools were enabled.
+    """
+    from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
+        ModelResponseIterator,
+    )
+
+    litellm_logging = MagicMock()
+
+    chunk = {
+        "candidates": [
+            {
+                "content": {
+                    "parts": [
+                        {
+                            "functionCall": {
+                                "name": "get_current_time",
+                                "args": {"timezone": "America/New_York"},
+                            },
+                            "thoughtSignature": "EsEDCr4DAdHtim...",  # Base64 signature
+                        }
+                    ]
+                },
+                "finishReason": "STOP",
+            }
+        ],
+        "usageMetadata": {
+            "promptTokenCount": 68,
+            "candidatesTokenCount": 120,
+            "totalTokenCount": 188,
+        },
+    }
+
+    iterator = ModelResponseIterator(
+        streaming_response=[], sync_stream=True, logging_obj=litellm_logging
+    )
+    streaming_chunk = iterator.chunk_parser(chunk)
+
+    # Verify that reasoning_content is present in the streaming delta
+    assert streaming_chunk.choices[0].delta.reasoning_content is not None
+
+    # Verify tool calls are also present
+    assert streaming_chunk.choices[0].delta.tool_calls is not None
+    assert len(streaming_chunk.choices[0].delta.tool_calls) == 1
+    assert streaming_chunk.choices[0].delta.tool_calls[0].function.name == "get_current_time"
+
+
 def test_check_finish_reason():
     finish_reason_mappings = VertexGeminiConfig.get_finish_reason_mapping()
     for k, v in finish_reason_mappings.items():