What does your AI feature cost per customer?
Put in your plan's price, how much light, normal and heavy customers use the feature, the tokens a use takes and the model's price. See the cost of one use, the gross margin of each group and overall, and what prompt caching, a cheaper model for most requests and a cap or an overage price do to it.
Your numbers are saved in this browser only. Nothing is sent to us.
Start from an example
Illustrations, not benchmarks. Your model vendor's usage export has the real token counts.
Your plan and customers
One use, and the models
Main model
List prices per million tokens, read on 15 October 2026. Check your vendor's page; they change, mostly down.
Cheaper model
For the requests a small model handles well enough.
Routing, cap and overage
Before you change the plan
Get the token counts from your vendor's usage export rather than guessing, and split customers by use, not by plan: the average hides the group that loses money. Test the cheaper model on fifty of your own requests before routing to it, and write what happens at a cap into the plan description before anyone hits it.
The worked example, what investors make of AI gross margin in 2026 and a monthly check are in our article when the AI feature eats the margin. If the fix is what you charge for, compare units with the billable unit comparison; if it is where the model runs, try the model hosting break-even calculator.
A model, not a forecast. More free tools: biz59.net/tools.