Skip to content
AI Coding Prices

Method

How we count

Tokens

Your prompt is counted in your browser with OpenAI's o200k tokenizer (the open-source js-tiktoken library). Nothing is uploaded: there is no server behind the calculator and no AI model reads your text. Until the tokenizer finishes loading, the count is a quick estimate of one token per four characters.

For OpenAI models the count is exact. Other providers use their own tokenizers, so for them it is an estimate. Anthropic, for example, says its tokenizer for Claude 4.7 and later produces about 30% more tokens for the same text than its older one.

Replies

A reply's length depends on the model and the task, so we show a range for a typical reply, thinking included. The effort setting picks the range:

These ranges are our own assumption, not provider figures. The big number on the ticket uses the middle of the range.

Prices

We use each provider's standard-tier API list price, taken from its own pricing page and dated. Where a long prompt moves to a higher rate, the calculator switches tiers at the provider's stated limit. Announced price rises take effect on their date in your browser. Batch, priority, regional and cached pricing are not included in the main figure.

Every price change is logged on the price changes page.

Subscription plans

Tell the calculator which plan you pay for and the ticket shows how much of it one prompt uses. No vendor publishes the size of a session or a usage pool, so be careful with these numbers:

Water per prompt

The water drop next to the headline shows how much water a prompt uses, as the provider itself has published it. Only three have published a figure, and each measures something different, so don't compare them with each other:

Providers with a greyed-out drop haven't published a per-prompt water figure. Nobody publishes water per token, so we don't scale these numbers to your prompt's length.

Sponsors

A sponsor can buy a clearly labelled slot under the results. Sponsors never change the results, the order of the table or what we write about a model.