O3-mini in assistants API: vector stores but no file search?

Am I missing something here ?
I can create an o3-mini assistant, attach a vector store but I cannot activate file_search ?
What is the point of having o3-mini in the assistants API if I can’t activate file_search ?

Hey,

I don’t think being able to attach a vector store means that file search is available for the model.

The behavior of o3-mini in the assistants API is the same in the chatgpt.com web UI.

+1 I dont understand that either.

@nikunj - can you comment? thanks!

Yes, tools (of which file_search is one) are not available on mini-o3.

Yeah be great if you could, im using gpt4o-mini with assistant and file_search, wanted to try the new o3-mini

Not sure if its never going to be available, or just not ready yet

The error message returned by the API seems pretty deliberate (although listing old models):

Error code: 400 - {'error': {'message': "The requested model 'o3-mini' cannot be used with the 'file_search' tool. Consider using 'gpt-4-turbo-preview', 'gpt-3.5-turbo', or a later model.", 'type': 'invalid_request_error', 'param': 'model', 'code': 'unsupported_model'}}

Currently, the model has issues with reliable function calling when a long context of turns is placed, so that may be why it is temporarily unavailable.

Otherwise, a model that supports tools->function->namespace should also support sending tools->file_search


What I’d like to see added to Assistants, given the current poor performance:

parameter

file_search_developer_instructions

  • Allows you to completely replace the existing instructions given to the AI model about how to use file_search (or myfiles_browser on some models).

*Important: You must understand and start from these default instructions used by OpenAI, making just the modifications to suit your application and inform of the knowledge behind a file search. {default_instructions}

2 Likes