Maybe I'm wrong about the fees. I was thinking that sending from a single address would use a smaller transaction than sending from multiple addresses.
Yes, you are wrong. The data size of the transaction has nothing to do with the addresses being used. It is all to do with
the number of inputs and outputs. So, a transaction that uses 10 inputs from 1 address, will likely have the same data size as a transaction that uses 1 input from 10 addresses... they both have 10 inputs.
There will be some variation depending on compressed vs. uncompressed addresses and legacy vs. segwit, but in general it won't matter (from a data size perspective) if you used one address or 100 addresses.