That always seems to happen upon release of a new model. People look at the benchmarks, which look good, because the model was almost certainly optimised for that. Then they start actually using it, and after a week or two we get the real assessment that is almost always less impressive than the initial reactions.

Interesting, I generally use High for all models, even non Anthropic ones.

This was one of the few times I tried Max, and the problem was the code (it listed as a "hole") was directly adjacent to the problem area, and not especially complex.

Kind of like looking at a washing machine and telling the customer to be careful because the inlet pipe will pump water into an empty box.