ISSUE 25 · ME+MACHINE
People talk about AI models like there’s a single scoreboard. Find the model at the top, use it for everything, and you’re set.
But using the most powerful model for every task can be like driving a Ferrari to the grocery store. It’ll get you there, but you’re paying for performance that a grocery run doesn’t need.
Think about what you actually ask AI to do. Rewrite a text. Fix broken code. Read a long report. Sort thousands of receipts. Each job has a different definition of “good.”
For a text, you might want something short that sounds like you. For broken code, you want careful reasoning. For thousands of receipts, accuracy matters, but so do speed and cost.
That’s where a scoreboard can mislead you. A model’s strength on a difficult coding test might matter a lot when you’re fixing a bug. It might matter much less when you’re rewriting a text. If a faster, cheaper model gives you the result you need, more capability may add very little.
A simple way to choose is to weigh four variables:
Difficulty: Start with a faster, simpler model for routine tasks; move up when the task requires deeper reasoning or specialized knowledge.
Accuracy: Use a stronger model when mistakes are costly, the instructions are ambiguous, or the output needs careful verification.
Speed: Prefer a faster model for interactive work, high-volume processing, or tasks where a good-enough answer is sufficient.
Cost: Use the least expensive model that reliably meets your quality bar, especially when processing many requests.
In practice, start with the cheapest model that seems capable, test its output on a few representative examples, and upgrade only if it misses your accuracy or reasoning requirements. You can also use a two-step approach: let a fast model handle routine work and send uncertain or high-stakes cases to a stronger one.
One model may be enough for most of your day. But when it struggles, changing models can be more useful than endlessly rewriting your prompt.
Sometimes the extra horsepower is worth it. But sometimes you just need to get the groceries home.
me+machine.
THE 30-SECOND VERSION
Match the model to the job: rewriting a text, fixing code, and processing receipts require different strengths.
Weigh accuracy against speed and cost.
Test affordable models on a few representative examples before choosing.
When a model struggles, try a more capable one instead of repeatedly rewriting the prompt.