Comment by epolanski

For 99% of the tasks any model really would do, even GPT 5.4 from an year ago.

It's only for long running/very difficult cases that the differences are here, and they are small.