Python: fix(bedrock): report the requested model on non-streaming responses

python/packages/bedrock/agent_framework_bedrock/_chat_client.py

@@ -420,7 +420,7 @@ async def _stream() -> AsyncIterable[ChatResponseUpdate]:
 
         async def _get_response() -> ChatResponse:
             raw_response = await asyncio.to_thread(self._invoke_converse, request)
-            return self._process_converse_response(raw_response, options)
+            return self._process_converse_response(raw_response, options, model=request["modelId"])
 
         return _get_response()
 
@@ -758,9 +758,12 @@ def _redacted_reasoning_content(
         )
 
     def _process_converse_response(
-        self, response: dict[str, Any], options: Mapping[str, Any] | None = None
+        self, response: dict[str, Any], options: Mapping[str, Any] | None = None, *, model: str | None = None
     ) -> ChatResponse:
-        """Convert Bedrock Converse API response to ChatResponse."""
+        """Convert Bedrock Converse API response to ChatResponse.
+
+        Converse does not return the model id, so ``model`` should be the ``modelId`` the request was sent with.
+        """
         output = response.get("output") or {}
         message = output.get("message") or {}
         content_blocks = message.get("content") or []
@@ -770,7 +773,7 @@ def _process_converse_response(
         usage_details = self._parse_usage(usage_source)
         finish_reason = self._map_finish_reason(output.get("completionReason") or response.get("stopReason"))
         response_id = response.get("responseId") or message.get("id")
-        model = response.get("modelId") or output.get("modelId") or self.model
+        model = response.get("modelId") or output.get("modelId") or model or self.model
         return ChatResponse(
             response_id=response_id,
             messages=[chat_message],

python/packages/bedrock/tests/test_bedrock_client.py

@@ -103,6 +103,73 @@ async def test_get_response_invokes_bedrock_runtime() -> None:
     assert response.usage_details and response.usage_details["input_token_count"] == 10
 
 
+class _StubConverseWithoutModelId(_StubBedrockRuntime):
+    """Returns the real Converse response shape, which does not echo the model id."""
+
+    def converse(self, **kwargs: Any) -> dict[str, Any]:
+        self.calls.append(kwargs)
+        return {
+            "output": {"message": {"role": "assistant", "content": [{"text": "hi"}]}},
+            "stopReason": "end_turn",
+            "usage": {"inputTokens": 3, "outputTokens": 1, "totalTokens": 4},
+        }
+
+
+@pytest.mark.parametrize("client_model", ["amazon.titan-text", None])
+async def test_get_response_reports_the_requested_model(client_model: str | None) -> None:
+    """A per-call model override must be the model reported on the response, like the streaming path."""
+    stub = _StubConverseWithoutModelId()
+    client = BedrockChatClient(
+        model=client_model,
+        region="us-west-2",
+        client=stub,  # pyrefly: ignore[bad-argument-type] # ty: ignore[invalid-argument-type] # pyright: ignore[reportArgumentType]
+    )
+
+    response = await client.get_response(
+        messages=[Message(role="user", contents=[Content.from_text(text="hello")])],
+        options={"model": "us.openai.gpt-6-sol"},
+    )
+
+    assert stub.calls[0]["modelId"] == "us.openai.gpt-6-sol"
+    assert response.model == "us.openai.gpt-6-sol"
+
+
+class _StubConverseAndStreamWithoutModelId(_StubConverseWithoutModelId):
+    def converse_stream(self, **kwargs: Any) -> dict[str, Any]:
+        self.calls.append(kwargs)
+        return {
+            "stream": _StubEventStream([
+                {"messageStart": {"role": "assistant"}},
+                {"contentBlockDelta": {"delta": {"text": "hi"}, "contentBlockIndex": 0}},
+                {"contentBlockStop": {"contentBlockIndex": 0}},
+                {"messageStop": {"stopReason": "end_turn"}},
+            ])
+        }
+
+
+@pytest.mark.parametrize("stream", [False, True])
+async def test_streaming_and_non_streaming_report_the_same_model(stream: bool) -> None:
+    """Both paths must report the per-call model override, not the client default."""
+    stub = _StubConverseAndStreamWithoutModelId()
+    client = BedrockChatClient(
+        model="amazon.titan-text",
+        region="us-west-2",
+        client=stub,  # pyrefly: ignore[bad-argument-type] # ty: ignore[invalid-argument-type] # pyright: ignore[reportArgumentType]
+    )
+    messages = [Message(role="user", contents=[Content.from_text(text="hello")])]
+    options = {"model": "us.openai.gpt-6-sol"}
+
+    if stream:
+        response_stream = client.get_response(messages=messages, options=options, stream=True)
+        response = await response_stream.get_final_response()
+    else:
+        response = await client.get_response(messages=messages, options=options)
+
+    assert stub.calls[0]["modelId"] == "us.openai.gpt-6-sol"
+    assert response.text == "hi"
+    assert response.model == "us.openai.gpt-6-sol"
+
+
 async def test_stream_yields_updates_as_converse_stream_events_arrive() -> None:
     """stream=True should use ConverseStream and yield text, tool call, finish and usage updates per event."""
     stub = _StubBedrockStreamRuntime([