However, state-of-the-art models are nevertheless showing improved performance on “reasoning” tasks like math, physics, and coding.
Source
A trend is graded on whether the publisher keeps it, never as a hit or a miss (D6, D33).