I use GPT 5.5. Sometimes it does what you say. It certainly finds silly mistakes in my code better than I could. But frequently enough it makes catastrophic architectural mistakes in its own code.