Running the same code in custom invocation and printing v.size() gives 81 in C++17 and 9090927 in C++11, so you're iterating over significantly more values when submitting under C++11 per test case. As for why, there appears to be a difference in the implementation of pow, as k takes on the values 1, 11, 111 in C++17 and 1, 11, 110 in C++11. Not sure why that happens though, I think the actual implementation of pow is architecture dependent. In my opinion, it's easier to just not involve floating point computation if possible.