The problem with benchmarks is that they are end to end.

You go from one prompt to the final solution, whereas for many of us it is about the experience of iterative, multi turn working.