When using modern AI models like ChatGPT or Claude, you have two separate dials: which model you use, and how much effort that model puts in. The distinction between the two is the distinction between being smarter and trying harder.
Both help with most problems on their own. For example, a mathematics PhD student is both very smart at mathematics and able to spend months, even years, on one problem. But running the strongest model on maximum effort all the time has real costs. Claude and ChatGPT subscriptions give you a fixed amount of usage per week, so working on “max settings” burns through it fast, and you risk running out. A model on higher settings also takes longer to answer, which can be unnecessary delay if a task does not require it. It is worth knowing which dial actually helps for the task in front of you.
A worked example
Say you need the answer to a biology question. Who would you ask:
- A bachelor’s student in biology, who will spend a day on it
- A PhD student in biology, who will spend a few hours on it
- A professor of biology, who will spend less than an hour on it
It depends on the problem. If the answer is buried in a stack of papers, favor the bachelor’s student, who has all day to search. If the problem needs advanced reasoning, favor the professor. The PhD student is a good middle ground.
Mistakes work the same way. The professor makes fewer of them, and is better at recognizing the ones they do make when they check their work, but may not have time to check it thoroughly. The bachelor’s student makes more mistakes but has time to check exhaustively. Ask any of the three to critique a scientific paper, and the professor’s critiques run sharper, while the student turns up more critiques.
In this analogy, the person is the model, and the time they spend is the effort setting:
- A bachelor’s student with a day to spend: Claude Haiku or ChatGPT Luna, effort set to max
- A PhD student with a few hours: Claude Sonnet or ChatGPT Terra
- A professor with under an hour: Claude Opus or ChatGPT Sol (Claude also has Fable, one tier above Opus) set to medium or low effort
Two rules of thumb
Turn the effort up when a task is going to take a lot of checking, repetition, or trial and error to land on a good answer. This covers things like searching a folder of documents or emails for one answer, making many edits to a document, or checking a spreadsheet cell by cell for errors.
Choose a higher-class model when the main challenge is judgment or cleverness rather than persistence. This covers things like strategizing, or any task where the ideas coming back feel dull or shallow. Higher-class models also tolerate vaguer instructions, because they are better at reading between the lines.
If you find that you are routinely reaching the end of your week without spending all of your AI usage, we advise selecting the higher-powered models more often.
Model names and tiers change often, so check your tool’s current documentation for what is available on your plan.
