Big brain, small brain: choosing the right model for the job
The biggest, slowest model is not the right default for most questions. Here is how we teach children to notice when a job needs the big brain and when the small one will do just as well.
We ran the exact same question - "what's a good snack that uses only three ingredients" - through a fast, small model and a large, slow one side by side. The small one answered in about a second with a perfectly good suggestion. The big one took several times longer and returned something barely more detailed. A ten-year-old in the group asked, reasonably, why anyone would ever choose the slow one if the fast one did the same job.
It's a fair question, and the honest answer is: for most everyday questions, you shouldn't. The biggest model is not the correct default. It is just the one that feels safest to reach for, which is a different thing entirely.
Which AI model should I use?
Match the size of the model to the size of the actual problem. A small, fast model handles simple, well-defined questions - a snack idea, a quick fact check, a short rewrite - just as well as a large one, and does it faster and at lower cost. A large, slower model earns its cost on genuinely hard problems: multi-step reasoning, tricky comparisons, anything where getting it right matters more than getting it fast.
- Quick facts, simple rewrites, short suggestions - small model, almost always.
- A tricky multi-step maths problem, a nuanced comparison, a complex plan - big model, worth the wait.
- When unsure, start small. Only reach for the bigger model if the small one's answer genuinely falls short.
"Big brain, small brain" as a household phrase
We found that giving the two model sizes a physical nickname made the choice stick far better than any technical explanation did. "Small brain" for quick stuff, "big brain" for the genuinely hard stuff - children start asking "does this really need the big brain" completely unprompted within a week of hearing the phrase once.
The biggest model is usually the wrong default - not because it's worse, but because most questions were never that hard to begin with.
Deciding honestly, not automatically
The honest limit is that "small" and "big" are relative, and the line between a question that needs the big brain and one that doesn't moves as models improve - what needed the big model last year might not need it this year. The habit worth keeping permanently is not a fixed rule but the pause itself: asking, before reaching for the most powerful option, whether the question in front of you actually requires it. The final article in this level covers the other side of that same habit - what happens when the same context gets pasted in and paid for over and over, and why that repetition has its own fix.
Questions we get asked
Which AI model should I use for a task?
Match the model's size to the actual difficulty of the question. Simple, well-defined tasks - quick facts, short rewrites, basic suggestions - are handled just as well by a small, fast model as a large one, at lower cost and higher speed. Save the larger, slower model for genuinely hard, multi-step problems where getting it right matters more than getting it fast.
Is a bigger AI model always better than a smaller one?
No - for most everyday questions, a smaller, faster model produces an equally good answer at a fraction of the cost and wait time. The larger model earns its extra cost specifically on complex, multi-step problems; using it as the automatic default for simple questions wastes both time and money without improving the answer.
How do you teach a child to choose between AI models?
Run the same question through both a small, fast model and a larger, slower one side by side, then compare the answers honestly rather than assuming the slower one must be better. Giving the two options a simple nickname, like "small brain" and "big brain," helps the choice become an automatic habit rather than a technical decision.