Pick the Right Model, Pick the Right Effort

Most people get worse results from AI than they should, not because the tool is weak, but because they're using one setting for everything. Learn two dials, and you get faster answers on easy tasks and genuinely better ones on hard tasks, without burning through your usage limit.

Ten minutes to learn. Here's the value, then exactly how to use it.

What you get

Faster answers on everyday questions, because you're not waiting on a heavyweight model to think hard about something simple.

Better answers on genuinely hard problems, because you've saved the top setting for when it's actually needed instead of spreading it thin.

More usage before you hit your limit, because you're not paying full price, in time and tokens, for questions that didn't need it.

Less frustration, because a shallow answer to a hard question is usually a settings problem, not a model problem.

The two dials

Dial one: which model. Think of it as gears. AI tools typically offer a range, from a fast, light model built for quick, simple, high-volume work, up to a top-tier model built for deep reasoning and complex, long-form problems. The names change often, but the range is consistent: lighter and faster at one end, heavier and more capable at the other.

Dial two: how hard it thinks. Most current AI tools let you set how much the model deliberates before answering, usually from a low setting up to a maximum. The default suits most everyday work. Turning it down gets you speed on something simple. Turning it up gets you deeper reasoning on something that actually needs it, at the cost of time and more of your usage.

How to actually use them

Step 1: Match the model to the task, not to habit. Before you ask anything, size up the task. Simple and repetitive doesn't need your best model. Complex and consequential does.

Step 2: Start on a low effort setting and raise it only if the answer disappoints. Don't default to maximum out of caution. Raise the setting when the first answer is too shallow, not before.

Step 3: Save the top model and top effort for what actually earns it. Multi-document analysis, decisions with real consequences, problems nobody has clearly solved before. Everything else runs fine, and faster, one or two notches down.

Step 4: Recheck the lineup periodically. Both major AI vendors have renamed their models more than once this year. The two-dial idea, capability and effort, is what's stable. The specific names are not.

The one habit worth keeping

Running everything on the heaviest model at maximum effort feels safe, but it's the AI equivalent of taking the motorway to your own driveway. Match the tool to the task, and both your answers and your usage limit go further.

Model names and specific effort settings move fast. Check the current lineup before relying on this for anything technical.