After spending a significant portion of today attempting to replace GPT-4.1 with GPT-6-Luna in a production-style agent workflow, I’ve come away convinced that GPT-4.1 should not be deprecated until a true substitute exists. This is not about benchmark scores…