For general use-case, closed-weight frontier models dominate.

Open-weights excel primarily at coding.