Set StreamedResponse.model_name from later streamed chunk for Azure OpenAI with content filter (#2951)

Author: DouweM

pydantic_ai_slim/pydantic_ai/models/openai.py

@@ -586,7 +586,8 @@ async def _process_streamed_response(
                 'Streamed response ended without content or tool calls'
             )
 
-        # ChatCompletionChunk.model is required to be set, but Azure OpenAI omits it so we fall back to the model name set by the user.
+        # When using Azure OpenAI and a content filter is enabled, the first chunk will contain a `''` model name,
+        # so we set it from a later chunk in `OpenAIChatStreamedResponse`.
         model_name = first_chunk.model or self._model_name
 
         return OpenAIStreamedResponse(
@@ -1352,9 +1353,12 @@ async def _get_event_iterator(self) -> AsyncIterator[ModelResponseStreamEvent]:
         async for chunk in self._response:
             self._usage += _map_usage(chunk)
 
-            if chunk.id and self.provider_response_id is None:
+            if chunk.id:  # pragma: no branch
                 self.provider_response_id = chunk.id
 
+            if chunk.model:
+                self._model_name = chunk.model
+
             try:
                 choice = chunk.choices[0]
             except IndexError:

tests/models/test_openai.py

@@ -398,7 +398,7 @@ async def test_stream_text(allow_model_requests: None):
 
 async def test_stream_text_finish_reason(allow_model_requests: None):
     first_chunk = text_chunk('hello ')
-    # Test that we fall back to the model name set by the user if the model name is not set in the first chunk, like on Azure OpenAI.
+    # Test that we get the model name from a later chunk if it is not set on the first one, like on Azure OpenAI with content filter enabled.
     first_chunk.model = ''
     stream = [
         first_chunk,
@@ -421,7 +421,7 @@ async def test_stream_text_finish_reason(allow_model_requests: None):
                     ModelResponse(
                         parts=[TextPart(content='hello world.')],
                         usage=RequestUsage(input_tokens=6, output_tokens=3),
-                        model_name='gpt-4o',
+                        model_name='gpt-4o-123',
                         timestamp=IsDatetime(),
                         provider_name='openai',
                         provider_details={'finish_reason': 'stop'},