About The Magic of AI
This site exists because pricing an AI feature is unreasonably hard, and it should not be.
The information needed to answer "what will this cost us" is public. It is also scattered across two dozen vendor pages, expressed in four incompatible units, and rewritten without notice. Working it out by hand means opening a dozen tabs, converting per-character prices into per-minute prices, remembering that output tokens are billed at a different rate to input tokens, and then discovering that the number you produced was stale by the time the meeting started.
The Magic of AI does that work once a week and hands you the result: a maintained price index covering 101 individually sourced items, and fourteen calculators that turn those prices into an answer for your specific workload.
What is actually here
The price index. Every LLM, embedding, image, video, speech and GPU rental price we track, each one carrying a link to the vendor page it came from. See the price index.
The calculators. Fourteen of them, covering the questions that come up repeatedly: what an API feature costs at volume, whether prompt caching pays for itself, where self-hosting overtakes an API, whether to stuff a context window or build retrieval. They run entirely in your browser. Nothing you type is transmitted anywhere, because there is nowhere for it to go — the pages are static files.
The change log. When a vendor moves a price, it is recorded with a date and a link. Over time this becomes the thing that is hardest to find elsewhere: a history of what AI actually cost, and when it changed. See the price change log.
The guides. Twelve pieces on the mechanics behind the numbers — why output tokens cost several times what input tokens cost, how to estimate token volumes before you have any traffic, what a migration to a cheaper model really costs. See the guides.
Editorial position
We are not affiliated with any AI vendor. No vendor pays for placement, and no calculator is weighted to favour anyone. When a cheap model is the right answer we say so, and when the honest answer is "this is not worth optimising", we say that too.
We are also not neutral about accuracy. Every price on this site links to the vendor page it was read from, and the methodology page explains exactly how the index is maintained and where its known limitations are. If a figure here disagrees with a vendor's own page, the vendor is right and we want to know about it.
How the site is funded
Display advertising, served through Google AdSense. That is the entire business model. There are no affiliate links in the calculators, no sponsored placements in the price index, and no paid tier.
Advertising has an obvious failure mode — it creates a reason to write for clicks rather than for readers — so the arrangement here is deliberately constrained. Ads never appear inside a calculator's controls, never between a field and its result, and never in a position that could be mistaken for part of the tool. If you find one that does, tell us and it will be removed.
Who writes it
The site is edited by John M., who is based in the Netherlands and has spent the last several years working with LLM APIs and the budgets attached to them. Research and drafting are assisted by AI tooling — it would be strange to run a site about AI economics and pretend otherwise — but every figure is traced to a primary source, every calculation is tested against hand-computed values, and nothing is published without review. The methodology page sets out how that works in practice.
Corrections, disagreements and requests for calculators we have not built yet are all welcome: get in touch.