Sometimes GPT 4 is ok. I often (1 in 50 times probably) have to switch to o1. Seems like Sonnet 3.5 (new) is optimized to solve all task well, but when there is complex problems it have never seen before o1 does better.

Using o1 I often find that it does not change the code, so switching back to Sonnet and ask to make the code changes and it often works.

Build and published 3 iPhone apps the last month this way.