Provider usage
Tokens in and out per task, priced per million tokens for each model, split between the routine model and the hard-case model by the share you set.
Free tool
Enter the volume, the models and the review time for one AI workflow. You see the provider spend, the full monthly cost and the cost per accepted result at public list prices, updated as you type.
The starting values are the worked example in our guide to AI cost management: 1,000 supplier assessments a month. Replace them with your own workflow and the result updates as you type.
of which $0.02 is provider usage
List prices, before caching discounts and retries.
Add your work email and the plan opens on this page, built from your numbers. Print it or save it as a PDF for the budget owner.
| Routing | Provider a month | Per accepted result |
|---|
Provider cost if every task in this workflow ran on one model. A cheaper model only saves money if its results still pass review, so judge the split on cost per accepted result.
| Model | Tier | Per task | A month |
|---|
List prices on 7 September 2026: platform.openai.com, claude.com, together.ai.
The daily cap is a tenth of the monthly budget, so a runaway job is refused once it reaches that amount in a day.
It follows the method in our guide to AI cost management. The prices come from that guide and every other figure from your inputs.
Tokens in and out per task, priced per million tokens for each model, split between the routine model and the hard-case model by the share you set.
Review minutes per task at the cost of a review hour, plus this workflow's share of the gateway, hosting and monitoring.
The total divided by the results that pass review. Results that fail review still cost money, so they stay in the total.
Public list prices per million tokens, read from the providers' pricing pages on 7 September 2026 and listed in our guide to AI cost management. They change often and your contract may differ, so use them to see the spread between models.
It's part of what an accepted result costs. In the starting example review time is the largest line, so a cheaper model barely moves the total.
No. Cached input is priced lower by several providers, which lowers the provider line. Retries and revisions add to it. If you know how often tasks are repeated, add the extra attempts to the tasks a month.
One gateway for every AI call, with a budget per team and the cheapest model that passes each task. In one of our model trials, a cheaper model passed the task for $0.94 where the default model hit its $14 cap.
Open our summary in the assistant you use, or copy the prompt.