As per the File Input Methods page, registering GCS files with the Gemini Files API should not count towards the 20 GB project storage quota, and the storage is technically uncapped (because the data is stored in a GCS bucket, and only pulled by the Files API per request).
I have about 70 GB of media files in a GCS bucket that I need to perform inference on. When I register the files with the Files API, everything is completely fine. However, when I submit a batch job (using the Gemini Batch API) which includes references to my GCS media files, the storage quota spikes immediately occupying the entire 70 GB, for some reason. At this point, the entire 20 GB quota is exhausted and I cannot upload or submit any more batch jobs. I have to delete all the files held by the Files API to free up the space, register those files again, and then submit a batch job again.
This bug is in direct contradiction with the information mention in the docs and should be addressed with a high priority.
