The existing type checkers are slow. That's their reasoning for creating it. The type checking world in python is now reasonably well specified (and the type checkers are beginning to work more similarly), so I would expect (once ty is out of beta) you could replace your type checker with ty and just see a 20x speedup in CI and much much faster IDE diagnostics.

If it was going from 3s -> 1s then you'd probably be right, but on my work codebase pyright (which is the faster one!) takes 15 seconds (ty takes 0.8s) and takes a bit for errors to appear in the IDE. This is now fast enough for us (mypy was taking >30s, which was not fast enough), but if the project grew another order of magnitude (which it might) then it would probably be too slow.

It's the same reason ruff is great, linting the codebase is so so fast.