A small personal peeve: it doesn't make sense to say that Sudoku is NP-hard, as the linked post does (or NP-complete). First of all, this benchmark (as far as I can tell) uses only traditional 9x9 puzzles, of which there is a finite number. Problems with a finite input space are trivial from the point of view of complexity theory; in order to talk about any kind of completeness or hardness results, you need to generalize Sudoku by defining NxN problems in some way.

Moreover, once you define NxN Sudoku (the generalization is easy), you need to pick the problem for which you want to prove a hardness result. The decision problem is typically defined as "given an instance of a problem, determine if there is a solution". For Sudoku, you would need to decide if a partial grid can be completed to a full grid, i.e. if the given puzzle is solvable. This has been proven to be NP-complete. However, this is not how the puzzle is usually done, including in this benchmark: you are given a puzzle which is known to have a solution, and the solution is known to be unique. You then need to find the solution. It's been shown that such a Sudoku problem is equivalent to the unambiguous satisfiability problem (given a logical formula which has a unique variable assignment that satisfies it, find that assignment). Now the unambiguous satisfiability problem is not known to be NP-complete when phrased as a decision problem.

Anyway, don't just throw around terms like "NP-hard". :)

EDIT: edited "unique satisfiability" to "unambiguous satisfiability". I believe this is the correct terminology for the "promise" version of the problem which I described, and to which Sudoku is equivalent. (The unique satisfiability problem is "decide if a logical formula has exactly one solution", where you are not told if it has more than one, none, or just one).