That statement would make a lot more sense with a benchmark graph based on standardised tasks.