Speed in Sudoku is almost never about thinking faster. It's about not wasting passes over the grid, not repeating work you've already done, and not losing four minutes to a mistake you made two minutes ago.
Time a few of your own solves and the pattern is consistent. A small share goes on hard deductions. A much larger share goes on rescanning ground you've already covered, hunting for something you can't name, and recovering from errors. Improve those three and your times drop without you getting any cleverer.
The single highest-value habit. When you place a digit, it changes the candidates in up to twenty cells — and the next available placement is very often in one of them. So don't return to your systematic sweep. Immediately check the row, column and box of the digit you just placed.
Solvers who do this move in bursts: one placement unlocks four more in fifteen seconds. Then the chain dies and you go back to scanning. Solvers who don't do this pay full scanning cost for every single digit on the board.
Random scanning means covering the same box six times and another box never. Pick a route and keep it — boxes left to right and top to bottom, or digits 1 through 9 — so that a completed pass means something. Knowing you've genuinely exhausted a technique is what lets you move on to the next one with confidence instead of anxiously re-checking.
Two failure modes here, in opposite directions.
Marking too early. Filling in every candidate on move one, on an Easy puzzle, costs several minutes and buys nothing — crosshatching would have finished it. Scan first. Mark only when scanning stops producing.
Marking too late. Refusing to write candidates on a Hard grid and trying to hold six cells' possibilities in your head. You'll be slower and you'll make errors. The moment you catch yourself re-deriving the same cell's candidates for the third time, write them down.
A middle path that works well: mark candidates only in the units you're actively working, not the whole grid. Local marks, cleaned up as you place digits.
A hint reveals a digit but not the reasoning, so it gets you unstuck without teaching you anything. Used occasionally on a genuine wall, that's fine. Used routinely, it removes the exact moments where you'd otherwise have learned to see a new pattern. If you use one, it's worth going back afterwards to work out what deduction you'd missed.
An error placed early can survive twenty more moves before it produces a visible contradiction, and by then you have no idea which digit was wrong. That's a restart, and a restart costs more than any amount of careful checking would have.
Speed follows recognition, and recognition follows repetition on the right level. Play at a difficulty where you stall a couple of times per puzzle and then find the way through. Too easy and you're drilling something you already do automatically; too hard and you spend the session stuck rather than solving.
If you want a concrete target, the Daily Challenge is the same board for everyone, so tracking your own time on it week to week is a fair measure of progress in a way that comparing across random practice puzzles is not.