(I read the answer on StackOverflow and this seems like an idea to try.)
The min-cost flow algorithm that we provide in OR-tools does not take fixed costs into account.
I thought of a way where you would duplicate the arcs, but there is no way to force the algorithm to use the "fixed-cost" arc with one-unit capacity before the other arcs.
So there is actually no way that we could improve our algorithm to take into account fixed costs (I would really have loved to), unless P=NP. I leave the proof of the latter to the reader ;-)
By the way, can I ask what kind of problem you're tackling to have such an NP-hard min-cost flow?
Cheers, and thank you for using our tools.
BdB