I’m just starting to experiment with the batch generation for GPT image 1.5, and I’m trying to figure out the pricing. When I run a batch, my internal calculations on pricing based on the returned usage don’t line up with what I’m seeing in the usage for my account. It’s almost as if the batch discount of 50% hasn’t been applied. Is that the case? Is it that the usage panel shows the full amount and then, when my account is charged, it rectifies the batch difference? It’s also possible that I just am calculating wrong, but I really don’t think so.
Has anyone else tried to verify their pricing for batch image runs? Or, I guess, batch runs in general.
I just confirmed that Batch API pricing is working as expected. This likely points to an implementation issue. One common mix-up is batching multiple requests rather than using the Batch API. Or, feel free to share your code.
Thank you for answering. I’m just still not clear on what “working as expected” is. Is the discount applied prior to the display in usage? Does usage display actual charges, or is that later rectified? Sorry if I missed this in the docs somewhere, but I just can’t seem to find any reference to this.
One option is to go to the usage page and examine ‘Total spend’ grouped by ‘Line Item’ which will display the input and output costs for using the different Models (and in this case also the API endpoints) split by token type.