[mrshu]

Do you think a more messier math benchmark (in terms of how it is defined) might be more difficult for these models to get?