Hi, i have any problems when i am trying to use file_search and previous response id. Does anyone else have a such problem? Much tokens are being used in the input and i got errors after few calls with gpt 4o
When using the Responses endpoint, if your input context (including tool returns from file_search that don’t expire), grows larger than the model input context, you will get an API error unless you set the truncation parameter to “auto”.
Unless you like $0.30 an API call, I would manage your own input context instead of reusing a request ID, until a better input management solution is released.
With your own control of the input, you can decide, for example, to have only one file search tool call result remain in the history to continue answering about, or have it last only until a user output is generated.
1 Like