Last updated: 2026-06-22

Tokens per dollar explained

Tokens per dollar is the purchasing power of a currency unit expressed in language-model tokens: the number of tokens a fixed amount of money buys at a given price. Providers publish prices per one million tokens because a single token costs a tiny fraction of a cent, and the per-million figure is the more legible quote. Tokens per dollar is the inverse of that quote, restated so the quantity bought — rather than the price paid — is the headline number.

How it works

If a provider charges price $p$ per one million tokens, the cost of a single token is $p / 10^6$. The tokens one currency unit buys is the reciprocal:

For a budget $B$, the total tokens it buys scale linearly with the budget:

Both relations are simple inversions of the per-million quote. Because input and output tokens are billed at different rates, the calculation is applied to one rate at a time; running it once with the input price and once with the output price brackets the purchasing power of a budget.

Worked example

A model priced at $3 per million input tokens costs $3 / 10^6$ — three millionths of a dollar — per token. One dollar therefore buys:

A budget of $10 buys ten times that:

so roughly 3.33 million tokens. At a rough conversion of 0.75 English words per token, 3.33 million tokens is on the order of 2.5 million words — useful for translating an abstract budget into a concrete volume of text the model can read or write.

Variations

The same inversion applies to any per-million quote. A cheaper model at $1 per million yields one million tokens per dollar; a premium model at $15 yields about 66,667. Output rates, typically three to five times the input rate, produce proportionally fewer tokens per dollar. The currency unit is arbitrary: the formula holds for any currency in which the price is quoted, so a price per million in yen gives tokens per yen.

Application

Tokens per dollar is most useful for cross-model comparison. Two models with different per-million prices and different capabilities can be placed on a common axis — tokens delivered per unit of spend — which makes the cost of switching legible. It also converts a project budget into a token allowance, the quantity that downstream estimates of request volume and context size depend on. To turn the token figure into a per-request or monthly bill, see Token Cost Calculator and LLM Monthly Cost Calculator; published rates often exclude the discounts modelled by Prompt Caching Savings Calculator and Batch API Savings Calculator.