Before you choose a model for a task, price it. Pick the kind of task, how often it runs and how much text goes in and out, and see what it would cost each month on today’s leading models. Then see what changes when the routine share of that work goes to a cheaper model.

Your task

A token is roughly three-quarters of a word. Change the numbers to match your own task.

Monthly cost by model

What routing saves

Send hard work to one model and routine work to a cheaper one.

Routing

Price is only part of the decision. This tool uses list prices and says nothing about quality. Before choosing, run 50 to 100 real examples of the task through the models you are considering and keep the ones whose answers you would sign off on.

Prices as of 5 October 2026, standard short-context list prices per million tokens. Caching, batch and long-context pricing change the numbers and are not included. Jev is a decision-only model from the startup TypeSafe AI: it picks from fixed answers and does not write text, so it appears only for classify-or-route tasks. A month is 30 days.

How to read the numbers

The shortest bar is a starting point. A model that costs a tenth as much is a bargain only if it does your work well enough, and the way to find out is to run 50 to 100 real examples of the task through it. Those examples are worth keeping. Every time a model is repriced, retired or overtaken, they tell you in an afternoon whether the replacement still does the job.

This calculator goes with my article Build for the Change, Not the Model, about why the companies handling AI well build around what they own and rent the models.