This is confusing to me because in their blogpost they show model benchmarks and it spanks Fable pretty soundly in most tests.