GPT-5 hasn't landed for me yet, but this has been my thought process too. This seems like a moment potentially equivalent to when Google got lowest-common-denominator-ed, when it stopped respecting your query keywords and doing "smart" things. If GPT-5 in practice turns out to be similarly optimized for lowest common denominator usage at the cost of precise controls over models, that'll be the thing that'll finally get me properly using Claude and Gemini and local models regularly.