Comment by nielstron

It could... but as pointed out by other the significance is unclear and per-model results have even less samples than the benchmark average. So: maybe :)