We migrated from the gpt-4.0-mini model to the gpt-4.1-nano model and our Assistant API stopped working with the uploaded file.
We have an uploaded .json file and we are using the RAG approach to pull the necessary vendor information from.
After upgrading to the new gpt-4.1-nano model, the Assistant API stopped outputting vendor information and behaves as if the file does not exist. If I switch to the gpt-4.1-mini model, the file starts working again.
After comparing the event logs from the two models mini and nano I noticed that the nano model does not try to provide file citations with file_citation annotations.
Is this behavior something expected or is it a bug in the model?
An important aspect for us is the speed of the model (thus we decided to try nano) and the ability of this model to work with files.