Picking the model per task is by far the biggest cost lever. Most tasks —classifying, extracting, summarizing— are solved by a small model at a fraction of the price of the large one. The most expensive mistake is using the most powerful model for everything: reserve the large one for what genuinely needs it and route the rest to the small one.
It is the first of the 7 levers to cut your OpenAI bill and the one that moves the needle most.
Which model by task type
| Task | Model | Relative cost |
|---|---|---|
| Classify, tag, route | Small | Very low |
| Extract data from text | Small | Very low |
| Summarize, rewrite | Small / medium | Low |
| Nuanced writing, tone | Medium | Medium |
| Complex reasoning, hard code | Large | High |
The mistake of using the large one for everything
The most powerful model is also the most expensive per token, sometimes by an order of magnitude. If you use it to classify tickets or extract fields from an invoice, you pay advanced-reasoning prices for a task a small model nails. At volume, that difference is most of a bloated bill.
How to decide
Start from the task, not the model. Ask what minimum capability solves it well: if a small model gives the quality you need, the decision ends there. Step up only when the small model falls short in a measurable way, not by default.
How to route
The usual pattern is a cheap gate in front: a small classifier (or a few rules) looks at each request and sends it to the right model. The simple work goes to the small one; only what genuinely needs it escalates to the large one. That way you pay for the expensive model only on the fraction of requests that justify it.
Put a number on it
Before fixing the routing, measure how much the cost changes by model. Put your tokens and your volume into the token cost calculator and compare the same work on a large model and on a small one: you will see the monthly difference at once. All of this is part of the cost per task that e-ficient measures.
Frequently asked questions
Which AI model should I use for each task? A small one for classifying, extracting and summarizing; a medium one for nuanced writing; reserve the large one for complex reasoning or hard code. Start with the small one and step up only if quality falls short.
How much do you save by picking the right model? It is the highest-impact lever: routing simple tasks to a small model can cut the cost of those requests by 50 to 90%.
Does it combine with the other levers? Yes. On top of the well-chosen model you stack prompt caching and the batch API.
At e-ficient we measure the cost per task of companies that already run AI in production and tell you how much you would save with each lever at your real volume. The first audit is free if you start a plan with a three-month commitment and returns a diagnosis within 72 hours.