Comment by jrflo

I'm assuming they are just going off published benchmarks and not re-benchmarking other people's models