fix(gemini): propagate provider modelVersion onto model responses

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py

@@ -2434,7 +2434,8 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
             completion_response = GenerateContentResponseBody(**completion_response)
 
         ## GET MODEL ##
-        model_response.model = model
+        model_version: Final = completion_response.get("modelVersion")
+        model_response.model = model_version if isinstance(model_version, str) else model
 
         ## CHECK IF RESPONSE FLAGGED
         if "promptFeedback" in completion_response and "blockReason" in completion_response["promptFeedback"]:
@@ -3264,7 +3265,12 @@ class ModelResponseIterator:
 
             processed_chunk: Final = GenerateContentResponseBody(**chunk)
             response_id: Final = processed_chunk.get("responseId")
-            model_response = ModelResponseStream(choices=[], id=response_id)
+            chunk_model_version: Final = processed_chunk.get("modelVersion")
+            model_response = ModelResponseStream(
+                choices=[],
+                id=response_id,
+                model=chunk_model_version if isinstance(chunk_model_version, str) else None,
+            )
 
             # Check if prompt is blocked due to content filtering
             blocked_response: Final = VertexGeminiConfig._check_prompt_level_content_filter(

tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py

@@ -5836,3 +5836,65 @@ def test_supported_reasoning_efforts_still_map(model):
             drop_params=False,
         )
         assert "thinkingConfig" in result
+
+
+def _generate_content_body() -> dict:
+    return {
+        "candidates": [
+            {
+                "content": {"role": "model", "parts": [{"text": "hi"}]},
+                "finishReason": "STOP",
+                "index": 0,
+            }
+        ],
+        "usageMetadata": {
+            "promptTokenCount": 5,
+            "candidatesTokenCount": 7,
+            "totalTokenCount": 12,
+        },
+    }
+
+
+def test_generate_content_transform_uses_reported_model_version():
+    """The served modelVersion must win over the requested name so downstream
+    pricing sees what actually ran."""
+    import httpx
+
+    body = {**_generate_content_body(), "modelVersion": "gemini-x-served"}
+    response: Final = VertexGeminiConfig()._transform_google_generate_content_to_openai_model_response(
+        completion_response=body,
+        model_response=ModelResponse(),
+        model="gemini-x",
+        logging_obj=MagicMock(),
+        raw_response=httpx.Response(200, headers={}),
+    )
+
+    assert response.model == "gemini-x-served"
+
+
+def test_generate_content_transform_falls_back_to_requested_model():
+    import httpx
+
+    response: Final = VertexGeminiConfig()._transform_google_generate_content_to_openai_model_response(
+        completion_response=_generate_content_body(),
+        model_response=ModelResponse(),
+        model="gemini-x",
+        logging_obj=MagicMock(),
+        raw_response=httpx.Response(200, headers={}),
+    )
+
+    assert response.model == "gemini-x"
+
+
+def test_streaming_chunk_carries_model_version():
+    from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
+        ModelResponseIterator,
+    )
+
+    chunk = {**_generate_content_body(), "modelVersion": "gemini-x-served"}
+    iterator: Final = ModelResponseIterator(
+        streaming_response=[], sync_stream=True, logging_obj=MagicMock()
+    )
+    streaming_chunk: Final = iterator.chunk_parser(chunk)
+
+    assert streaming_chunk.model == "gemini-x-served"