5.6 came out with three different models and a wide range of possible effort-levels to choose from. How do you pick the right one?
Well, most of the time medium effort is the right choice: not too token-hungry, but good judgment. If you want more reasoning time (giving the model more time and compute to think through something) then increase the effort level to high.
Tasks that warrant high-effort: complex research, tricky debugging, complex multi-step work.
Extra-high and max, I generally don’t use unless I’ve tried high-effort first and the result wasn’t good enough.
Ultra only if I want multiple agents to work in parallel. If you’re dealing with a highly complex problem and want multiple agents to work on different parts of it, then ultra is the right effort level. Keep in mind, this burns through tokens.
Terra I only use when I am near my usage limits and want to keep working, and only for simple tasks. (E.g. add a table of contents to this website, proofread this article for spelling and grammar, change all CTA buttons to have rounded corners and bold font.)