diff --git a/litellm/llms/vertex_ai/gemini/transformation.py b/litellm/llms/vertex_ai/gemini/transformation.py
index 13e2238fdf6..f175bb5acdf 100644
--- a/litellm/llms/vertex_ai/gemini/transformation.py
+++ b/litellm/llms/vertex_ai/gemini/transformation.py
@@ -882,8 +882,7 @@ def _gemini_convert_messages_with_history(
                 _message_content = assistant_msg.get("content", None)
                 reasoning_content = assistant_msg.get("reasoning_content", None)
                 thinking_blocks = assistant_msg.get("thinking_blocks")
-                if reasoning_content is not None:
-                    assistant_content.append(PartType(thought=True, text=reasoning_content))
+                emitted_thinking = False
                 if thinking_blocks is not None:
                     for block in thinking_blocks:
                         if block["type"] == "thinking":
@@ -904,6 +903,9 @@ def _gemini_convert_messages_with_history(
                                             text=block_thinking_str,
                                         )
                                     )
+                                emitted_thinking = True
+                if not emitted_thinking and reasoning_content is not None:
+                    assistant_content.append(PartType(thought=True, text=reasoning_content))
                 if _message_content is not None and isinstance(_message_content, list):
                     _parts = []
                     for element in _message_content:
diff --git a/tests/llm_translation/test_prompt_factory.py b/tests/llm_translation/test_prompt_factory.py
index 7b03736920b..e5b207e9137 100644
--- a/tests/llm_translation/test_prompt_factory.py
+++ b/tests/llm_translation/test_prompt_factory.py
@@ -2490,3 +2490,109 @@ def test_has_tool_with_name_anthropic_shape_without_type_field():
 def test_has_tool_with_name_not_a_list():
     assert not has_tool_with_name(None, "my_tool")
     assert not has_tool_with_name("not a list", "my_tool")
+
+
+def test_gemini_no_duplicate_reasoning_parts_when_both_fields_present():
+    """
+    Regression (#37973): when an assistant message has both reasoning_content and
+    thinking_blocks, only one reasoning part must be emitted. Signed thinking_blocks
+    take priority; reasoning_content is a fallback used only when thinking_blocks is absent.
+    """
+    reasoning_text = "Two plus two is four."
+    signature = "test-signature-abc123"
+
+    messages = [
+        {"role": "user", "content": "What is 2+2?"},
+        {
+            "role": "assistant",
+            "content": "4",
+            "reasoning_content": reasoning_text,
+            "thinking_blocks": [
+                {
+                    "type": "thinking",
+                    "thinking": reasoning_text,
+                    "signature": signature,
+                }
+            ],
+        },
+        {"role": "user", "content": "Multiply by 3."},
+    ]
+
+    contents = _gemini_convert_messages_with_history(messages=messages)
+    assistant_turn = contents[1]
+    assert assistant_turn["role"] == "model"
+
+    parts = assistant_turn["parts"]
+
+    # Only the signed block should appear — no extra plain thought part
+    signed_parts = [p for p in parts if p.get("thoughtSignature") is not None]
+    plain_thought_parts = [p for p in parts if p.get("thought") is True]
+
+    assert len(signed_parts) == 1, (
+        f"Expected 1 signed reasoning part, got {len(signed_parts)}: {signed_parts}"
+    )
+    assert len(plain_thought_parts) == 0, (
+        f"reasoning_content must not be emitted when thinking_blocks is present; "
+        f"got {len(plain_thought_parts)} extra plain thought part(s)"
+    )
+    assert signed_parts[0]["thoughtSignature"] == signature
+
+
+def test_gemini_reasoning_content_fallback_when_no_thinking_blocks():
+    """
+    When an assistant message has reasoning_content but no thinking_blocks,
+    a plain thought part should still be emitted (the fallback path still works).
+    """
+    reasoning_text = "I reasoned about this carefully."
+
+    messages = [
+        {"role": "user", "content": "Hello."},
+        {
+            "role": "assistant",
+            "content": "Hi there.",
+            "reasoning_content": reasoning_text,
+        },
+        {"role": "user", "content": "How are you?"},
+    ]
+
+    contents = _gemini_convert_messages_with_history(messages=messages)
+    assistant_turn = contents[1]
+    assert assistant_turn["role"] == "model"
+
+    parts = assistant_turn["parts"]
+
+    plain_thought_parts = [p for p in parts if p.get("thought") is True]
+    assert len(plain_thought_parts) == 1, (
+        f"Expected 1 plain thought part from reasoning_content fallback, got {len(plain_thought_parts)}"
+    )
+    assert plain_thought_parts[0]["text"] == reasoning_text
+
+
+def test_gemini_reasoning_content_fallback_when_thinking_blocks_empty():
+    """
+    Greptile review: when thinking_blocks is non-null but empty (or has no
+    usable signed blocks), reasoning_content must still be emitted as fallback.
+    """
+    reasoning_text = "I reasoned carefully."
+
+    messages = [
+        {"role": "user", "content": "Hello."},
+        {
+            "role": "assistant",
+            "content": "Hi.",
+            "reasoning_content": reasoning_text,
+            "thinking_blocks": [],  # non-null but empty — no usable blocks
+        },
+        {"role": "user", "content": "How are you?"},
+    ]
+
+    contents = _gemini_convert_messages_with_history(messages=messages)
+    assistant_turn = contents[1]
+    parts = assistant_turn["parts"]
+
+    plain_thought_parts = [p for p in parts if p.get("thought") is True]
+    assert len(plain_thought_parts) == 1, (
+        f"reasoning_content fallback must fire when thinking_blocks is empty; "
+        f"got {len(plain_thought_parts)} plain thought part(s)"
+    )
+    assert plain_thought_parts[0]["text"] == reasoning_text
\ No newline at end of file
diff --git a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py
index 4679b978f78..34280872ad6 100644
--- a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py
+++ b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py
@@ -2781,3 +2781,95 @@ def test_gemini_server_side_tool_signature_not_duplicated_on_text():
     assert "thoughtSignature" not in text_part
     tool_call_part = next(p for p in parts if "toolCall" in p)
     assert tool_call_part["thoughtSignature"] == "server_side_signature"
+
+
+def test_gemini_no_duplicate_reasoning_parts_when_both_fields_present():
+    """
+    Regression (#37973): when an assistant message has both reasoning_content and
+    thinking_blocks, only one reasoning part must be emitted. Signed thinking_blocks
+    take priority; reasoning_content is a fallback used only when thinking_blocks is absent.
+    """
+    reasoning_text = "Two plus two is four."
+    signature = "test-signature-abc123"
+
+    messages = [
+        {"role": "user", "content": "What is 2+2?"},
+        {
+            "role": "assistant",
+            "content": "4",
+            "reasoning_content": reasoning_text,
+            "thinking_blocks": [
+                {
+                    "type": "thinking",
+                    "thinking": reasoning_text,
+                    "signature": signature,
+                }
+            ],
+        },
+        {"role": "user", "content": "Multiply by 3."},
+    ]
+
+    contents = _gemini_convert_messages_with_history(messages=messages)
+    parts = contents[1]["parts"]
+
+    signed_parts = [p for p in parts if p.get("thoughtSignature") is not None]
+    plain_thought_parts = [p for p in parts if p.get("thought") is True]
+
+    assert len(signed_parts) == 1, (
+        f"Expected 1 signed reasoning part, got {len(signed_parts)}: {signed_parts}"
+    )
+    assert len(plain_thought_parts) == 0, (
+        f"reasoning_content must not be emitted when thinking_blocks is present; "
+        f"got {len(plain_thought_parts)} extra plain thought part(s)"
+    )
+    assert signed_parts[0]["thoughtSignature"] == signature
+
+
+def test_gemini_reasoning_content_fallback_when_no_thinking_blocks():
+    """
+    When an assistant message has reasoning_content but no thinking_blocks,
+    a plain thought part should be emitted as fallback.
+    """
+    reasoning_text = "I reasoned about this carefully."
+
+    messages = [
+        {"role": "user", "content": "Hello."},
+        {
+            "role": "assistant",
+            "content": "Hi there.",
+            "reasoning_content": reasoning_text,
+        },
+        {"role": "user", "content": "How are you?"},
+    ]
+
+    contents = _gemini_convert_messages_with_history(messages=messages)
+    parts = contents[1]["parts"]
+
+    plain_thought_parts = [p for p in parts if p.get("thought") is True]
+    assert len(plain_thought_parts) == 1
+    assert plain_thought_parts[0]["text"] == reasoning_text
+
+
+def test_gemini_reasoning_content_fallback_when_thinking_blocks_empty():
+    """
+    When thinking_blocks is non-null but empty, reasoning_content must still be emitted.
+    """
+    reasoning_text = "I reasoned carefully."
+
+    messages = [
+        {"role": "user", "content": "Hello."},
+        {
+            "role": "assistant",
+            "content": "Hi.",
+            "reasoning_content": reasoning_text,
+            "thinking_blocks": [],
+        },
+        {"role": "user", "content": "How are you?"},
+    ]
+
+    contents = _gemini_convert_messages_with_history(messages=messages)
+    parts = contents[1]["parts"]
+
+    plain_thought_parts = [p for p in parts if p.get("thought") is True]
+    assert len(plain_thought_parts) == 1
+    assert plain_thought_parts[0]["text"] == reasoning_text