How this site is built

Price Per Token answers one question: what a specific prompt costs across every model you might send it to. Everything below is how that number is produced, so you can decide how much to trust it.

Where prices come from

Rates are pulled from the LiteLLM pricing dataset, an open, community-maintained record of published API rates. The sync runs as a script, not by hand, so a model added upstream appears here without anyone remembering to add it. The date at the bottom of every page is the last successful sync — currently 2026-08-11.

Figures are list prices per million tokens. They exclude batch discounts, negotiated enterprise rates, free tiers, and anything a reseller might charge on top. Where a provider bills differently past a context threshold, that second rate is shown too — 28 of the 246 models here have one, and quoting only the base rate would understate long prompts by up to 2×.

Which token counts are exact

Only OpenAI publishes the tokenizer its models bill against. Counts for those 45 models are produced by running that encoder on your text, in your browser. Every other family — Anthropic, Google, Mistral, and the rest — has no public tokenizer, so their counts are approximations derived from published ratios.

Those two cases look identical on most comparison sites. Here every model carries an exact or estimated badge, because a cost estimate built on a silent guess is worse than one you know to double-check.

What is not tracked

Your prompt never leaves your browser. Tokenizing happens client-side and nothing is uploaded, logged, or stored. See the privacy page for the full picture.

Corrections

Prices change constantly and upstream data occasionally lags a provider's own announcement. If a figure here disagrees with a provider's pricing page, the provider is right — tell us and it gets fixed in the next sync.