fix(bigmodel): forward Codex's include and parallel_tool_calls, keep the OpenAI org header off BigModel, and pass unchanged Messages stream events through

examples/bigmodel.md

@@ -56,8 +56,6 @@ GLM-5.3, GLM-5.3-Flash, and GLM-5.3-FlashX use always-on thinking. GLM-5.3 suppo
 
 Set `store=True` when creating a Responses request if you need continuation, retrieval, or deletion. Use the response ID returned by LiteLLM. BigModel documents a seven-day retention period and does not support cancel, compact, native WebSocket, or file_search
 
-Codex sends `include` and `parallel_tool_calls` on every Responses request, and LiteLLM does not forward either to BigModel. Set `drop_params: true` on the deployment so LiteLLM drops them instead of answering 400
-
 The model catalog and packaged backup include GLM-5.3, GLM-5.3-Flash, and GLM-5.3-FlashX. Prices are expressed in USD/token using a static 7.0 CNY/USD conversion, with the source and verification date in each entry's metadata. Use existing `litellm_params` cost fields for contract-specific deployment prices
 
 API references: [Chat Completions](https://docs.bigmodel.cn/cn/guide/develop/openai/introduction), [Responses](https://docs.bigmodel.cn/cn/guide/develop/responses/introduction), [Messages](https://docs.bigmodel.cn/cn/guide/develop/claude/introduction), and [pricing](https://docs.bigmodel.cn/cn/guide/start/pricing), checked on 2026-10-27

litellm/llms/bigmodel/messages/transformation.py

@@ -62,7 +62,7 @@ def _normalized_data(data: str) -> str:
         if body.get("type") == "message_start" and isinstance(message, dict)
         else _normalize_message(body)
     )
-    return json.dumps(normalized)
+    return data if normalized == body else json.dumps(normalized, ensure_ascii=False)
 
 
 def _normalize_event(event: ServerSentEvent) -> bytes:

litellm/llms/bigmodel/responses/transformation.py

@@ -32,6 +32,8 @@ _SUPPORTED_PARAMS: Final = (
     "max_output_tokens",
     "tools",
     "tool_choice",
+    "parallel_tool_calls",
+    "include",
     "reasoning",
     "text",
     "prompt_cache_key",
@@ -62,6 +64,7 @@ class _RequestedTools(BaseModel):
 
     tools: JsonValue = None
     tool_choice: JsonValue = None
+    parallel_tool_calls: JsonValue = None
 
 
 def _requested_tools(logging_obj: Logging) -> _RequestedTools:
@@ -97,6 +100,7 @@ def _normalize_response(
     output: Final = body.get("output")
     tools: Final = body.get("tools")
     tool_choice: Final = body.get("tool_choice")
+    parallel_tool_calls: Final = body.get("parallel_tool_calls")
     input_tokens: Final = usage.get("input_tokens") if isinstance(usage, dict) else None
     output_tokens: Final = usage.get("output_tokens") if isinstance(usage, dict) else None
     normalized_usage: Final = (
@@ -108,8 +112,10 @@ def _normalize_response(
         else usage
     )
     normalized: Final[JsonValue] = {
-        "parallel_tool_calls": False,
         **body,
+        "parallel_tool_calls": (
+            requested.parallel_tool_calls is True if parallel_tool_calls is None else parallel_tool_calls
+        ),
         "tools": (requested.tools or []) if tools is None else tools,
         "tool_choice": (requested.tool_choice or "auto") if tool_choice is None else tool_choice,
         **({"usage": normalized_usage} if "usage" in body else {}),

litellm/main.py

@@ -2672,8 +2672,7 @@ def _complete_custom_openai(
         )
     organization = (
         organization
-        or litellm.organization
-        or get_secret("OPENAI_ORGANIZATION")
+        or (None if custom_llm_provider == "bigmodel" else litellm.organization or get_secret("OPENAI_ORGANIZATION"))
         or None  # default - https://github.com/openai/openai-python/blob/284c1799070c723c6a553337134148a7ab088dd8/openai/util.py#L105
     )
     openai.organization = organization

tests/unit/llms/bigmodel/test_bigmodel_provider.py

@@ -180,6 +180,23 @@ async def test_chat_uses_native_endpoint_and_maps_token_limit(is_async: bool) ->
         assert json.loads(route.calls.last.request.content)["thinking"] == {"type": "enabled"}
 
 
+@respx.mock
+def test_chat_does_not_send_the_openai_organization_to_bigmodel(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setenv("OPENAI_ORGANIZATION", "org-openai-only")
+    route: Final = respx.post(f"{BASE}/api/paas/v4/chat/completions").respond(
+        json={
+            "id": "chat_test",
+            "object": "chat.completion",
+            "created": 1,
+            "model": MODEL,
+            "choices": [{"index": 0, "message": {"role": "assistant", "content": "Hello"}, "finish_reason": "stop"}],
+            "usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2},
+        }
+    )
+    _ = litellm.completion(model=f"bigmodel/{MODEL}", messages=[{"role": "user", "content": "Hello"}])
+    assert "openai-organization" not in route.calls.last.request.headers
+
+
 @pytest.mark.parametrize("is_async", (False, True))
 @pytest.mark.asyncio
 @respx.mock
@@ -451,23 +468,37 @@ def test_responses_failure_envelope_raises_its_status_instead_of_a_parse_error(
 
 @pytest.mark.parametrize("stream", (False, True))
 @respx.mock
-def test_responses_without_tools_in_the_answer_echo_the_requested_tools(stream: bool) -> None:
+def test_responses_forward_codex_tool_params_and_echo_them_when_the_answer_omits_them(stream: bool) -> None:
     tools: Final = [{"type": "function", "name": "weather", "parameters": {"type": "object", "properties": {}}}]
-    completed: Final = {"type": "response.completed", "sequence_number": 0, "response": response_payload()}
-    respx.post(f"{BASE}/api/v1/responses").respond(
+    completed: Final = {
+        "type": "response.completed",
+        "sequence_number": 0,
+        "response": {**response_payload(), "parallel_tool_calls": None},
+    }
+    route: Final = respx.post(f"{BASE}/api/v1/responses").respond(
         **(
             {"content": f"event: response.completed\ndata: {json.dumps(completed)}\n\n"}
             if stream
-            else {"json": response_payload()}
+            else {"json": {**response_payload(), "parallel_tool_calls": None}}
         )
     )
     answer: Final = litellm.responses(
-        model=f"bigmodel/{MODEL}", input="Hello", tools=tools, tool_choice="required", stream=stream
+        model=f"bigmodel/{MODEL}",
+        input="Hello",
+        tools=tools,
+        tool_choice="required",
+        parallel_tool_calls=True,
+        include=["reasoning.encrypted_content"],
+        stream=stream,
     )
     response: Final = tuple(answer)[-1].response if stream else answer
     validated: Final = OpenAIResponse.model_validate(response.model_dump())
+    sent: Final = json.loads(route.calls.last.request.content)
+    assert sent["parallel_tool_calls"] is True
+    assert sent["include"] == ["reasoning.encrypted_content"]
     assert validated.tools[0].name == "weather"
     assert validated.tool_choice == "required"
+    assert validated.parallel_tool_calls is True
 
 
 @pytest.mark.asyncio
@@ -820,6 +851,9 @@ async def test_messages_stream_preserves_native_events() -> None:
         "web_search_requests": 0,
         "web_fetch_requests": 0,
     }
+    assert tuple(line for line in lines if line.startswith("data: {"))[1:] == tuple(
+        f"data: {json.dumps(event, ensure_ascii=False)}" for event in events[1:]
+    )
     assert "event: ping" in lines
     assert "data: [DONE]" in lines
     assert b"id: bigmodel-event\n" in output[0]