I actually noticed this too. GPT 5.5 is much more "hands on" with calling tools to debug issues and verify results. I did all my tests in Cursor but I don't know if they use a different system prompt for each model.