AI token cost calculator

Estimate inference charges using token counts and prices you enter. The calculator includes no built-in provider pricing, so it cannot silently rely on an outdated rate.

Before you enter your numbers

Enter measured input and output tokens for a representative request.
Convert quoted prices to dollars per million tokens before entering them.
Include retries in request volume when billed, and model other billing categories separately.

A worked comparison

At illustrative rates of $1 per million input tokens and $5 per million output tokens, 2,000 input plus 500 output tokens cost $0.0045 per request. Ten thousand identical requests cost $45. Doubling only output to 1,000 tokens raises that total to $70. These rates are invented examples, not a provider offer.

Keep these assumptions with your result

Workload and model identifier
Pricing source and date
Measured input and billable output
Request volume including billed retries
Other meters excluded
Estimate period and later invoice reconciliation

Common questions

Why can a single request cost less than a cent?

A per-million rate applied to a small request can produce a sub-cent amount. The result retains extra decimal places for small nonzero totals so it is not displayed as free.

Does the estimate include cached or reasoning tokens?

It has only the two meters you enter. If your provider bills additional categories or reports billable output differently from visible output, use its documented usage and add the missing categories separately.

How the tool works

Cost = requests × (input tokens × input price + output tokens × output price) ÷ 1,000,000. The illustrative starting values produce $20 input cost and $25 output cost, totaling $45.

What the result does not tell you

Example prices are invented, not provider quotes. Cached tokens, reasoning tokens, tool calls, images, audio, retries, storage, taxes and batch discounts are excluded. Token counts vary by model and content; use measured usage for a better estimate.

Use the estimate responsibly

Replace the example values with records that cover the same period and scope. Save the input assumptions with your result so that another person can reproduce it. Change one assumption at a time to see why the result moves. Do not treat more decimal places as evidence that an estimate is more certain.