Getting faster at mental math is a training problem, not a talent problem. The same two levers that build any skill — the right difficulty and consistent short practice — apply here too. This guide is the practical version: what to actually do, this week.
If every answer feels obvious, the level is too easy to build anything. If you're guessing, it's too hard to learn from. The sweet spot is a level where you're right most of the time but occasionally miss — that's where speed actually improves.
Math Games' Stats page shows your accuracy for exactly this reason: if you're above 95%, move up a level; if you're below 75%, drop one.
Most people aren't slow at everything — they're slow at one thing, usually division or subtraction with borrowing, and it drags down their overall feel for the whole game.
Check accuracy by operation on your Stats page. If division is the outlier, spend two or three rounds on it deliberately, rather than hoping it improves as a side effect of general practice.