Which AI model to use. You don't need the strongest one for a spreadsheet

Someone switches on the strongest model at the highest thinking level to turn a list of addresses into a table. They wait while it thinks, and get the same table a lighter model would have produced right away. A slice of this week's usage limit is gone as well.

Which AI model to use depends on the task. The strongest model isn't the best choice for everything, and the highest thinking level often changes nothing.

Which AI model to use for which job

ChatGPT and Claude both offer several models. In Claude today they are Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, but these names change fast, so it's easier to think in categories. Anthropic's documentation compares them plainly. The most capable one is the slowest, and the lightest one is the fastest.

The strongest model pays off for deep analysis, a hard problem without an obvious answer, or a long document where you need to connect parts that are far apart. Comparing two versions of a long specification is a good example, or looking for contradictions in a large tender offer.

A lighter model is enough for a table, an email or reformatting a document. It does the job just as well and faster. There's nothing to analyse deeply there, it only has to be rewritten correctly.

Thinking mode and effort levels

The second dial is how much the model thinks before it answers. In Claude, the model menu next to the send button has an Effort option with Low, Medium, High, Extra high and Max. One level is marked as the default for each model. On the newest Claude models thinking is always on, and this level sets how deep it goes.

Anthropic is straightforward about it. Higher effort gives more thorough answers, but it takes longer and uses more, so you reach your limit sooner. Low and Medium work well for routine tasks. For simple questions and everyday writing the defaults are enough, and Max is meant for work that needs the deepest reasoning.

In ChatGPT the model picker has a fast Instant mode for everyday questions and a Thinking mode that takes longer. What exactly you get depends on your plan.

More thinking helps where there is something to think about. For an email or a table it usually doesn't change the result, it only makes you wait longer.

Usage limits on subscription plans

Paid Claude plans have two kinds of usage limits. A session limit resets every five hours, and weekly limits reset once a week. How fast you use them depends on conversation length, the features you use, the model and the effort level. Work in the browser, in the desktop app and in Claude Code all counts toward the same limit.

You can see how much you've used in Settings, under Usage. There are progress bars for the current session and for the week.

The most capable models use up the limit faster, and on some plans they aren't part of the plan's limits at all and are billed separately. If someone sets them for everything, the limit can run out halfway through the week.

In ChatGPT, the limits for Thinking also depend on the plan.

A simple rule for every day

Start with the default setting. If an answer is clearly too shallow, raise the level or switch to a stronger model in the same chat. In Claude the change applies from the next response.

For simple, repetitive tasks go the other way. A lighter model or a lower effort level gives you a quick answer and leaves more of the limit for work that really needs it.

If the same simple tasks come back every day, like copying orders from emails into a spreadsheet, it is often better to set them up once as an automation than to open a chat each time.

Keep the strongest model and the highest level for tasks that really need thinking. For everything else a lighter setting is enough, and the limit stays for what matters.