Because people, for some stupid reasons, still read and make decisions by these public benchmarks, which makes zero sense. Create your own benchmarks and evaluations, then use those, it's pretty the only way to get any qualitative answers to how well a model will work for you and your user cases.