Comment by Aurornis

It’s not a cost savings relative to paying API prices even if you’re suspending it.

This is an option if you must run local inference, you’re not sensitive to speed, and the budget is low.

It’s not going to be cheaper than paying API prices for the model though.