GPT-5 unable to run in LangSmith Playground

LangSmith Product Help Observability & Evals
1

I was trying to test GPT-5 in the Playground when it was unable to provide an output due to top_p being an unsupported parameter in the request.

Error message

BadRequestError(‘Error code: 400 - {\‘error\’: {\‘message\’: “Unsupported parameter: \‘top_p\’ is not supported with this model.”, \‘type\’: \‘invalid_request_error\’, \‘param\’: \‘top_p\’, \‘code\’: \‘unsupported_parameter\’}}’)Traceback (most recent call last): File “/usr/lib/python3.11/site-packages/langchain_core/runnables/base.py”, line 2383, in _atransform_stream_with_config File “/usr/lib/python3.11/site-packages/langchain_core/runnables/base.py”, line 3446, in _atransform File “/usr/lib/python3.11/site-packages/langchain_core/runnables/base.py”, line 5721, in atransform File “/usr/lib/python3.11/site-packages/langchain_core/runnables/base.py”, line 1578, in atransform File “/usr/lib/python3.11/site-packages/langchain_core/language_models/chat_models.py”, line 645, in astream File “/usr/lib/python3.11/site-packages/langchain_openai/chat_models/base.py”, line 2822, in _astream File “/usr/lib/python3.11/site-packages/langchain_openai/chat_models/base.py”, line 1370, in _astream File “/usr/lib/python3.11/site-packages/openai/resources/chat/completions/completions.py”, line 2603, in create File “/usr/lib/python3.11/site-packages/openai/_base_client.py”, line 1794, in post File “/usr/lib/python3.11/site-packages/openai/_base_client.py”, line 1594, in request openai.BadRequestError: Error code: 400 - {‘error’: {‘message’: “Unsupported parameter: ‘top_p’ is not supported with this model.”, ‘type’: ‘invalid_request_error’, ‘param’: ‘top_p’, ‘code’: ‘unsupported_parameter’}}

I saw another forum post that experienced a similar error due to temperature being unsupported by GPT-5. Is there a way to disable all of the unsupported parameters of GPT-5 so that it can be used in LangSmith?

2

Hey @evnpoon,

I’m guessing you’re opening a previously created LLM run with a different OpenAI model?

If you open the prompt settings gear here:

image
image636×380 12.3 KB

Are you able to set an alternate config as a stopgap?

Jacob

3

Hi @jacoblee93,

You’re right – I was opening a previously created LLM run with a different OpenAI model.

I tried opening the prompt settings and changing the config to some pre-saved one rather than an unsaved one. This seemed to do the trick – I was able to change the model of the pre-saved config to GPT-5 and run it to get an output with Top P still enabled.

Would you be able to provide some context as to why this is the solution? Why is it that the unsaved config cannot run GPT-5 while a pre-saved one that was changed to use this model could?

4

Hey @evpoon,

It’s because we also pass params from your previously run naively into the new model. Will flag and try to get some smoother fix planned - perhaps a warning saying some config will be dropped.

Jacob

5

Got it – thanks!