Openai has been focusing a lot on cutting down overthinking is the feel I get. If you look at the artificial analysis tokens per task benchmark Sol especially at lower effort uses far less tokens than the competition.