You’ve gotta be living under a rock if you dont think the models themselves have been improving over the last year, lol.

We are bumping into a log scale problem where people arent fully grasping how big of a difference going from an x% error rate to a y% error rate is in actual practice for where it matters.