New guidance for GPT-5.6 models advocates outcome-first prompting, showing major cost and performance gains by defining goals instead of step-by-step instructions.
