GPT-5 w/ minimal reasoning and gpt-5-chat-latest doesn't support vector stores / file search

API
1

We need a model that is both intelligent, fast and in the GPT-5 family that can support file search.

2

gpt-5-chat model is like chatgpt-4o-latest on the API: it is the model that ChatGPT uses - writing lots of markdown and chatty patterns with emoji.

It uses no reasoning. It does not take a reasoning parameter, along with other unaccepted parameters (but will accept temperature).

It is also not provided with any tool call ability and has lower usage limits, merely for your experimentation.

“gpt-5” is your model name. The Responses API is required for internal tools.

3

GPT-5 with minimal reasoning doesn’t support File Search either (testing in the Playground)

4

You are correct, of course:

openai.BadRequestError: Error code: 400 - {'error': {'message': "The following tools cannot be used with reasoning.effort 'minimal': file_search.", 'type': 'invalid_request_error', 'param': 'tools', 'code': None}}

The API shutting internal functions off with minimal reasoning in playground makes this a deliberate decision.

The GPT-5 documentation section written about “minimal” could call out this significant deviation from expectations, but mention is absent.

The minimal setting performs especially well in coding and instruction following scenarios,

How the is it going to do that coding with no code interpreter?

The basis for this may be that internal tool calling is contained within reasoning, and thinking generation can continue again after a result. We see this intent with the “oopsie” release of web_search on o4-mini that was clawed back from API developers until it was implemented as desired.

You can prompt your way into an approximation of “minimal”. To keep the AI from exploring on my dime to make a test, for example I added: Budget: Use a single tool call and single query per user question.

a minimal reasoning on a file search result with “low”:

ResponseUsage(input_tokens=7207, input_tokens_details=InputTokensDetails(cached_tokens=2640), output_tokens=371, output_tokens_details=OutputTokensDetails(reasoning_tokens=128), total_tokens=7578)

by:

“\n\nThinking/pre-reasoning: disabled. Immediately close reasoning and begin tool call or user output.”