Contact
Corrections, requests and questions about methodology all go to the same place, and pricing corrections get priority.
This is a one-person project, so replies are not instant — but reported pricing errors are usually verified and fixed within a few days.
Reporting a pricing or token-count error
This is the most valuable thing you can email about. Every price on the site is transcribed by hand from a provider's rate card, and providers change prices without announcement, so the table drifts out of date between reviews.
To make a correction quick to action, please include:
- The model or platform affected.
- The figure shown on the site and the figure you believe is correct.
- A link to the provider's own pricing page showing the correct rate. This matters more than anything else — a correction without a source takes far longer to verify.
For token-count discrepancies, it also helps enormously to include the exact text you tokenized, the count the site gave, and the count the provider's own tokenizer or API usage response reported. Remember that only the OpenAI models are counted exactly; everything else is a labelled approximation, so a few percent of difference is expected rather than a bug. A discrepancy above roughly 10% is worth reporting.
Requesting a model or platform
New models get added regularly. When requesting one, a link to its official pricing page and its context window size is all that is needed to get it into the table.
Two things affect whether a request makes it in. Models need a publicly published per-token price — anything priced only by private enterprise agreement cannot be compared meaningfully. And models need enough real-world use to be worth a row; the table is more useful short and accurate than exhaustive.
Gateway and inference platform requests are welcome on the same basis: a link to public pricing, and a sense of what the platform is genuinely better at than the ones already listed.
Questions about the methodology
If you want to know why a specific number is what it is, the methodology page documents the approximation ratios, the multi-turn and prompt-caching formulas, and the known limitations. It is written to be specific enough to check the arithmetic yourself.
If something there is unclear, wrong, or you think a modelling choice is mistaken, that is worth an email. Disagreements about methodology have previously changed how the tool calculates things.
Privacy, advertising and legal
For questions about what data the site collects, see the privacy policy — the short answer is that text you paste never leaves your browser, and the only third-party code is Google AdSense. For data protection requests, or anything relating to the terms of use, email the address above and put "Privacy" or "Legal" in the subject line.
What this project cannot help with
To save you the wait for a reply: this site is a calculator, not a consultancy. It cannot help with debugging your API integration, choosing a model for a specific task (that requires evaluating output quality, which the site does not attempt), negotiating enterprise pricing with a provider, or billing disputes — those need the provider's own support channel, since they hold your account and usage data.
Advertising, sponsorship and paid-placement enquiries are declined. The model and gateway tables contain no affiliate links and no paid positions, and keeping it that way is what makes the comparison worth reading.