openai[patch]: type reasoning_effort (#28825)

libs/partners/openai/langchain_openai/chat_models/base.py

@@ -454,6 +454,16 @@ class BaseChatOpenAI(BaseChatModel):
     """Total probability mass of tokens to consider at each step."""
     max_tokens: Optional[int] = Field(default=None)
     """Maximum number of tokens to generate."""
+    reasoning_effort: Optional[str] = None
+    """Constrains effort on reasoning for reasoning models. 
+    
+    o1 models only.
+
+    Currently supported values are low, medium, and high. Reducing reasoning effort 
+    can result in faster responses and fewer tokens used on reasoning in a response.
+    
+    .. versionadded:: 0.2.14
+    """
     tiktoken_model_name: Optional[str] = None
     """The model name to pass to tiktoken when using this class. 
     Tiktoken is used to count the number of tokens in documents to constrain 
@@ -599,6 +609,7 @@ def _default_params(self) -> Dict[str, Any]:
             "stop": self.stop or None,  # also exclude empty list for this
             "max_tokens": self.max_tokens,
             "extra_body": self.extra_body,
+            "reasoning_effort": self.reasoning_effort,
         }
 
         params = {

libs/partners/openai/pyproject.toml

@@ -24,7 +24,7 @@ ignore_missing_imports = true
 [tool.poetry.dependencies]
 python = ">=3.9,<4.0"
 langchain-core = "^0.3.27"
-openai = "^1.55.3"
+openai = "^1.58.1"
 tiktoken = ">=0.7,<1"
 
 [tool.ruff.lint]

libs/partners/openai/tests/integration_tests/chat_models/test_base.py

@@ -1089,19 +1089,13 @@ async def test_astream_response_format() -> None:
         pass
 
 
-def test_o1_max_tokens() -> None:
-    response = ChatOpenAI(model="o1-mini", max_tokens=10).invoke("how are you")  # type: ignore[call-arg]
-    assert isinstance(response, AIMessage)
-
-    response = ChatOpenAI(model="gpt-4o", max_completion_tokens=10).invoke(
-        "how are you"
-    )
-    assert isinstance(response, AIMessage)
-
-
-def test_developer_message() -> None:
-    llm = ChatOpenAI(model="o1", max_tokens=10)  # type: ignore[call-arg]
-    response = llm.invoke(
+@pytest.mark.parametrize("use_max_completion_tokens", [True, False])
+def test_o1(use_max_completion_tokens: bool) -> None:
+    if use_max_completion_tokens:
+        kwargs: dict = {"max_completion_tokens": 10}
+    else:
+        kwargs = {"max_tokens": 10}
+    response = ChatOpenAI(model="o1", reasoning_effort="low", **kwargs).invoke(
         [
             {"role": "developer", "content": "respond in all caps"},
             {"role": "user", "content": "HOW ARE YOU"},

libs/partners/openai/tests/unit_tests/chat_models/test_base.py

@@ -882,3 +882,9 @@ def test__get_request_payload() -> None:
     }
     payload = llm._get_request_payload(messages)
     assert payload == expected
+
+
+def test_init_o1() -> None:
+    with pytest.warns(None) as record:  # type: ignore[call-overload]
+        ChatOpenAI(model="o1", reasoning_effort="medium")
+    assert len(record) == 0