What is AI actually costing you?

AI providers bill by the token — roughly ¾ of a word — separately for what you send, what the model writes back, and the hidden "thinking" reasoning models do in between. Describe your usage below to see your real run rate and five controls that bring it down.

Spent while you've been reading this

$0.0000

at your current run rate

$0/month

1 · Your usage

Support or FAQ bot with a system prompt and short answers — adjust the numbers to match your reality.

Requests per day Every message, API call, or agent step counts Avg tokens sent per request Instructions + documents + conversation history Avg tokens received per request Avg reasoning tokens per request Hidden "thinking" by reasoning models and agents — billed at output rates. Set 0 if unsure. Model tier you mostly use

Usage growth per month10%

Most teams grow 5–15%/mo once AI catches on

2 · Cost controls

Simple requests moved to a cheaper model

FAQ answers and routine tasks rarely need your best model

Repeated context you stop re-sending

Providers discount cached instructions by ~90%

Responses made shorter

Output tokens cost ~5x input — verbosity is expensive

Thinking effort reduced on simple tasks

Reasoning models "think" at output prices — most simple tasks don't need it

Non-urgent work moved to batch

Overnight jobs typically get 50% off

Current monthly cost

With controls applied

Recovered per year

Where the money goes today

What each control saves per month

The next 12 months

Assumptions: 30.4 days/month · controls applied in sequence (cheaper models → caching → shorter responses → thinking effort → batch) so savings never double-count · reasoning tokens billed at output rates · caching modeled at 85% hit rate, hits ~10% of list price, misses ~125% (write premium) · batch = 50% discount · default pricing reflects mid-2026 market rates — edit the pricing table to match your provider's current rate card. Estimates only; validate against your actual invoices.