By skyforbes been noticing this a lot lately: a prompt that feels rock solid on one model just crumbles when u move it to a different one. same wording, same structure, same logic… completely different behavior. sometimes it’s tiny stuff like tone drifting, but other times the whole reasoning pattern collapses.
i’m starting to think it’s less about “good phrasing” and more about how different models interpret hierarchy, constraints, and task flow. like one model respects boundaries perfectly, another blends everything into one blob unless u separate layers. i read in god of prompt consistency setups where the same framework behaves totally differently across models unless the rules are isolated cleanly. anyone else run into this? is it just model personality, or are there deeper architectural differences that make prompts “non-portable” across systems?