logprobs and refusal validation error for non-OpenAI model providers #1616
Description
Describe the question
When running an agent, it responses fail validation because logprobs is missing from streamed responses.
This seems related to PR #1226 / #1246, which was merged after July 28, 2025,
Debug information
- openai-agents== 0.2.9
- openai==1.101.0
- Python 3.11
Repro steps
session = SQLiteSession("conversation_123")
agent = Agent(
name="Assistant",
instructions="Reply very concisely.",
model=OpenAIResponsesModel(
model="openai/gpt-oss-20b",
openai_client=AsyncOpenAI(
base_url="http://0.0.0.0:8000/v1",
api_key="not-needed",
),
),
)
result = await Runner.run(
agent,
"What city is the Golden Gate Bridge in?",
session=session
)
print(result.final_output) # "San Francisco"
result = await Runner.run(
agent,
"What state is it in?",
session=session
)
print(result.final_output) # "California"
result = Runner.run_sync(
agent,
"What's the population?",
session=session
)
print(result.final_output)Error Trace:
Error getting response: Error code: 400 - {
'error': {
'message': "3 validation errors for ValidatorIterator
0.ResponseOutputTextParam.logprobs
Input should be iterable [type=iterable_type, input_value=None, input_type=NoneType]
0.ResponseOutputRefusalParam.refusal
Field required [type=missing, input_value={'annotations': [], 'text': '...'}, input_type=dict]
0.ResponseOutputRefusalParam.type
Input should be 'refusal' [type=literal_error, input_value='output_text', input_type=str]",
'type': 'BadRequestError',
'code': 400
}
}
Expected behavior
logprobs should default to [] (as fixed in PR #1246), so missing fields don’t cause validation errors.