Quick or considered: two families

In the dropdown menu of your interface, the models fall into two families. This is the most useful distinction, and the least explained one.

The fast ones

They respond within a second or two. They produce their answer word by word, with no preliminary step. For rewording a message, summarising a text, translating, answering a general knowledge question: they are perfect, and the extra thinking would add nothing.

The thoughtful ones

Before answering, they produce an intermediate reasoning, often invisible, where they break down the problem, try things, and correct themselves. This takes ten, thirty seconds, sometimes more. You will often see a note like "thinking in progress" flash by.

On a multi-step problem, a calculation, a contradiction to untangle, the gap is clear. On a rewording, it is nil, and you have simply waited.

The simple rule

Does the problem have several steps? If yes, go for the thoughtful one. Otherwise, the fast one.

The trap of habit

Many people always pick "the most powerful" one, out of reflex. It is often slower, costlier, and no better: on a simple task, a thoughtful model can even over-analyse a straightforward request and produce a convoluted answer.