Looks like this table becomes less relevant, as there are models on the top who trained on test datasets and spoiled the leaderboard.