Skip to content

The agent engineering platform.

openai[patch]: add explicit attribute for service tier (#31005)

ccurme629b7a5a43d0
diff --git a/libs/langchain/tests/unit_tests/chat_models/test_base.py b/libs/langchain/tests/unit_tests/chat_models/test_base.py
index 3ed6bc4717..9c24390de4 100644
--- a/libs/langchain/tests/unit_tests/chat_models/test_base.py
+++ b/libs/langchain/tests/unit_tests/chat_models/test_base.py
@@ -119,6 +119,7 @@ def test_configurable() -> None:
"reasoning_effort": None,
"frequency_penalty": None,
"seed": None,
+ "service_tier": None,
"logprobs": None,
"top_logprobs": None,
"logit_bias": None,
diff --git a/libs/partners/openai/langchain_openai/chat_models/base.py b/libs/partners/openai/langchain_openai/chat_models/base.py
index 6a5dc7a599..9e740f38a9 100644
--- a/libs/partners/openai/langchain_openai/chat_models/base.py
+++ b/libs/partners/openai/langchain_openai/chat_models/base.py
@@ -538,6 +538,10 @@ class BaseChatOpenAI(BaseChatModel):
However this does not prevent a user from directly passed in the parameter during
invocation.
"""
+ service_tier: Optional[str] = None
+ """Latency tier for request. Options are 'auto', 'default', or 'flex'. Relevant
+ for users of OpenAI's scale tier service.
+ """
use_responses_api: Optional[bool] = None
"""Whether to use the Responses API instead of the Chat API.
@@ -655,6 +659,7 @@ class BaseChatOpenAI(BaseChatModel):
"n": self.n,
"temperature": self.temperature,
"reasoning_effort": self.reasoning_effort,
+ "service_tier": self.service_tier,
}
params = {
diff --git a/libs/partners/openai/tests/integration_tests/chat_models/test_base.py b/libs/partners/openai/tests/integration_tests/chat_models/test_base.py
index c36c739028..c0cbba20f0 100644
--- a/libs/partners/openai/tests/integration_tests/chat_models/test_base.py
+++ b/libs/partners/openai/tests/integration_tests/chat_models/test_base.py
@@ -215,12 +215,15 @@ async def test_openai_abatch_tags(use_responses_api: bool) -> None:
assert isinstance(token.text(), str)
[email protected](retries=3, delay=1)
def test_openai_invoke() -> None:
"""Test invoke tokens from ChatOpenAI."""
- llm = ChatOpenAI(max_tokens=MAX_TOKEN_COUNT) # type: ignore[call-arg]
+ llm = ChatOpenAI(
+ model="o4-mini",
+ service_tier="flex", # Also test service_tier
+ )
- result = llm.invoke("I'm Pickle Rick", config=dict(tags=["foo"]))
+ result = llm.invoke("Hello", config=dict(tags=["foo"]))
assert isinstance(result.content, str)
# assert no response headers if include_response_headers is not set
diff --git a/libs/partners/openai/tests/unit_tests/chat_models/test_base.py b/libs/partners/openai/tests/unit_tests/chat_models/test_base.py
index a212d82540..e114ec0687 100644
--- a/libs/partners/openai/tests/unit_tests/chat_models/test_base.py
+++ b/libs/partners/openai/tests/unit_tests/chat_models/test_base.py
@@ -1732,3 +1732,9 @@ def test__construct_responses_api_input_multiple_message_types() -> None:
# assert no mutation has occurred
assert messages_copy == messages
+
+
+def test_service_tier() -> None:
+ llm = ChatOpenAI(model="o4-mini", service_tier="flex")
+ payload = llm._get_request_payload([HumanMessage("Hello")])
+ assert payload["service_tier"] == "flex"