Lesson 5 of 7
Right-size the model
6 min read
The same job can cost 20× more on one model than another. So why would anyone pay for the expensive one — and when should you?
The biggest lever
Model tiers span a wide price range — often 5–25× from the lightest to the flagship. Bigger isn't always better for a given task: summarising, sorting, and simple replies rarely need the top model. Right-sizing means starting with the cheapest model that still passes your quality bar, and only moving up when it genuinely fails.
Choosing the right-sized model is the biggest single cost lever.
Two discounts on top
Once the model is right, two mechanisms trim more. [Prompt caching](glossary://prompt-caching) reuses an already-processed chunk of your prompt instead of re-reading it, cutting repeated-input cost. [Batch processing](glossary://batch-processing) runs non-urgent jobs together for roughly half price. Neither matters if you're overpaying for the wrong-sized model first — sizing comes before discounts.
Caching and batching help — but only after you've right-sized the model.
Default to the cheapest capable model; reserve the flagship for the hard cases that actually need it, and route the rest cheaply.
Cutting the bill, in order
- —Right-size first: cheapest model that passes.
- —Prompt caching cuts repeated-input cost.
- —Batch non-urgent jobs for ~50% off.
What's the biggest lever for cutting an AI bill?
Continue in the app
Take the whole What AI Costs course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.