Which Model For Which Job

Start Here — The Whole Answer

Here from the video? This is the rule, then the why is underneath.

First, Stop Worrying About New Releases

When a new model appears in your dropdown at work, the honest answer is that it barely matters for you. The frontier models have been clever enough for everyday professional work for a good while now — the writing, the analysis, the spreadsheets, the summaries. A model that's a generation or two old is still far smarter than the job you're giving it, and that stays true for whatever comes out next.

The question that actually changes your results isn't which generation you're on. It's which size you pick for each job — and that's a decision nobody at work explains, because most people don't know it's a decision at all.

The Firm

Every provider sells the same ladder — a big deep thinker, a middle workhorse, and a small fast one. In Claude the ladder is named Opus, Sonnet, Haiku. In ChatGPT and Copilot the names change with each release, but the sizes are always there, and the same logic applies. Easiest way to hold it in your head is people at a firm.

Why The Expensive One Is Often Cheaper

This is the counterintuitive bit. Yes, the big model costs more each time you use it. But give a big, complicated job to the mid tier worker and they do what a mid level person does — they spend ages trying to figure it out, go round in loops planning the best approach, and sometimes come back with "sorry, I couldn't finish that". You wait longer, you retry, you tidy up after it.

The senior partner just gets on with it and comes back right the first time. So on anything genuinely hard, the expensive model tends to be cheaper and faster overall — fewer attempts, less of your time, a better answer. That's what people keep finding with the current generation, and it's why my honest recommendation is to make the big model your daily driver.

And if you're on a subscription — a work Copilot licence, a Claude or ChatGPT plan — you're not paying per message anyway. Using the senior partner costs you nothing extra. Use it.

The Honest Bit

Two places this rule bends. If you're paying per message — a business running AI at scale through an API, thousands of calls a day — the maths genuinely changes, and the smaller models earn their keep on the routine work. That's a whole topic of its own. And if you're a cutting edge developer or researcher, new releases really do move things for you — this page is for the rest of us.

One more thing no model fixes — a vague ask. The senior partner with a bad brief is just a slower wrong answer. Which model you pick is the second most important decision; what you actually tell it is the first. There's a full guide on that in context engineering.

Make It Your Rule

Paste this into whichever AI you use — Claude, ChatGPT, Copilot, any of them. It'll interview you about your actual work and hand you back a personal version of this rule.

I want a simple personal rule for choosing which AI model to use for each task. Ask me about the kinds of work I do in a typical week, one question at a time, no more than four questions. Then, using the idea that models come in sizes like people at a firm (a senior partner, a mid tier worker, a junior), tell me which of my regular tasks belongs with which size, and give me the rule as one sentence I can remember without writing it down.

Where This Goes Next