I feel like people analyzing LLM models and acting as if they know what they are talking about is kind of the new "fools gold". As an end-user can you actually prove that what you are experiencing is because of the model change, or its just a different circumstance and you got a different reply? Also, it feels counterintuitive to have an artificial intelligence do crazy calculations and write code, but having to manually select which model to use for it. Best would be, if there was a model in the first place that would just choose the most optimal model to execute your request, otherwise why bother automating everything in the first place?