Benchmarks compare open models against closed products, not closed models. We might be missing what were actually paying for