Most probably the last version they had access to for benchmarking, or otherwise.
I'm assuming they are just going off published benchmarks and not re-benchmarking other people's models
I'm assuming they are just going off published benchmarks and not re-benchmarking other people's models