fix: streaming container_id and consistent Pydantic types in output

- Populate container_id on streaming code_interpreter_results by
  re-emitting at message_delta when container info arrives
- Reconstruct Pydantic OutputCodeInterpreterCall objects from plain
  dicts in _extract_tool_result_output_items so responses_output
  has uniform types across streaming and non-streaming paths

litellm/llms/anthropic/chat/handler.py

@@ -546,6 +546,7 @@ class ModelResponseIterator:
         self._server_tool_inputs: Dict[str, Any] = {}
         self.tool_results: List[Dict[str, Any]] = []
         self._current_server_tool_id: Optional[str] = None
+        self._container_id: Optional[str] = None
 
     def check_empty_tool_call_args(self) -> bool:
         """
@@ -726,7 +727,7 @@ class ModelResponseIterator:
                     type="code_interpreter_call",
                     id=call_id,
                     code=code,
-                    container_id=None,
+                    container_id=self._container_id,
                     status="completed",
                     outputs=log_outputs,
                 )
@@ -928,6 +929,17 @@ class ModelResponseIterator:
                 finish_reason, usage, container = self._handle_message_delta(chunk)
                 if container:
                     provider_specific_fields["container"] = container
+                    # Store container_id and re-emit code_interpreter_results
+                    # so stream_chunk_builder's last-value-wins picks up the
+                    # version with container_id populated.
+                    container_id = (
+                        container.get("id") if isinstance(container, dict) else None
+                    )
+                    if container_id and self.tool_results:
+                        self._container_id = container_id
+                        provider_specific_fields["code_interpreter_results"] = (
+                            self._build_code_interpreter_results()
+                        )
             elif type_chunk == "message_start":
                 """
                 Anthropic

litellm/responses/litellm_completion_transformation/transformation.py

@@ -1738,10 +1738,7 @@ class LiteLLMCompletionResponsesConfig:
             )
         )
         if tool_result_items:
-            result_by_id = {
-                (item.get("id") if isinstance(item, dict) else item.id): item
-                for item in tool_result_items
-            }
+            result_by_id = {item.id: item for item in tool_result_items}
             replaced_ids = set(result_by_id.keys())
             responses_output = [
                 (
@@ -1778,7 +1775,14 @@ class LiteLLMCompletionResponsesConfig:
                 continue
             results = psf.get("code_interpreter_results")
             if results and isinstance(results, list):
-                output_items.extend(results)
+                for item in results:
+                    # In the streaming path, items are plain dicts after
+                    # model_dump() in stream_chunk_builder.  Reconstruct
+                    # Pydantic objects so responses_output has a uniform type.
+                    if isinstance(item, dict):
+                        output_items.append(OutputCodeInterpreterCall(**item))
+                    else:
+                        output_items.append(item)
         return output_items
 
     @staticmethod

tests/test_litellm/llms/anthropic/chat/test_code_interpreter_results_extraction.py

@@ -58,7 +58,8 @@ def test_extract_tool_result_output_items_from_pydantic_objects():
 
 
 def test_extract_tool_result_output_items_from_dicts():
-    """Streaming path: after model_dump(), code_interpreter_results are plain dicts."""
+    """Streaming path: after model_dump(), code_interpreter_results are plain dicts.
+    _extract_tool_result_output_items reconstructs them as Pydantic objects."""
     items = [
         {
             "type": "code_interpreter_call",
@@ -72,7 +73,8 @@ def test_extract_tool_result_output_items_from_dicts():
     resp = _make_model_response(code_interpreter_results=items)
     result = LiteLLMCompletionResponsesConfig._extract_tool_result_output_items(resp)
     assert len(result) == 1
-    assert result[0]["id"] == "srvtoolu_01AAA"
+    assert isinstance(result[0], OutputCodeInterpreterCall)
+    assert result[0].id == "srvtoolu_01AAA"
 
 
 def test_extract_tool_result_output_items_empty():
@@ -258,8 +260,9 @@ def test_end_to_end_streaming_chunks_to_code_interpreter_output():
     )
     assert len(tool_result_items) == 1
     item = tool_result_items[0]
-    # Items are dicts after the model_dump path
-    assert item["type"] == "code_interpreter_call"
-    assert item["id"] == "srvtoolu_01AAA"
-    assert item["code"] == "echo e2e_test"
-    assert item["outputs"][0]["logs"] == "e2e_test\n"
+    # Items are reconstructed as Pydantic OutputCodeInterpreterCall objects
+    assert isinstance(item, OutputCodeInterpreterCall)
+    assert item.type == "code_interpreter_call"
+    assert item.id == "srvtoolu_01AAA"
+    assert item.code == "echo e2e_test"
+    assert item.outputs[0].logs == "e2e_test\n"