I tend to build mip problems that take days to make a good approximation (using gurobi). Those can not be modeled using pulp, because I run out of memory.

I’ll check out python-mip, thanks for the tip.