- Methodology
How we source and check prices
Every figure on this site is a published list price taken from the vendor's own pricing page. Nothing is estimated, modelled, or supplied by a vendor in exchange for placement.
List prices checked
Where the numbers come from
The dataset holds 77 cloud services and 25 AI models from 53 providers. Each row records the rate, the URL of the pricing page it was read from, and the month it was read — currently August 2026. That source link appears under every table, so any figure can be checked against the vendor in one click.
Where a vendor does not publish a rate for something, the field is left empty and the page shows a dash. It is never filled with a zero or an estimate. This matters more than it sounds: a missing rate treated as zero would make that provider look cheapest in every comparison it appeared in.
How the monthly costs are calculated
Rate cards alone do not answer “which is cheaper”, so each provider's published rates are applied to a small set of fixed workload profiles — the same profiles on every page, so any two comparisons can be read against each other. Every line in a cost breakdown shows its own arithmetic.
Where a provider bills on an axis it does not publish a rate for, the estimate is marked incomplete and sorted to the bottom of the table rather than shown as a low number. An incomplete estimate is a floor, not a figure, and the pages say so.
What the estimates deliberately exclude
- Committed-use, reserved-instance, and enterprise-agreement discounts, which routinely cut large bills by 30–60%.
- Regional price variation. Rates are the cheapest common region unless stated otherwise; some regions cost substantially more.
- Per-request and API operation charges, which matter for object storage with small files.
- Support plans, which at the hyperscalers are often a percentage of total spend.
- Minimum retention periods and early-deletion fees on cold storage tiers.
- Taxes, and currency conversion where a provider does not bill in USD.
The comparison this site makes is list price against list price, for identical usage. That is a useful shortlist, not a quotation. Before committing to a provider, price your actual workload with the vendor.
How often prices are rechecked
Every row carries the month it was last verified, shown next to the price on every page. AI model pricing moves fastest and is checked most often; cloud infrastructure rates change more slowly. When a row goes stale the date on the page says so rather than the page pretending otherwise.
The site is a static build with no database and no live scraping, which means the prices you see are exactly the prices in the repository at the last deploy. There is no state in which a page renders a number that was never reviewed.
Commercial independence
Rankings are computed from published prices by code in the repository, and no vendor can buy a position. That is enforced structurally rather than promised: the ranking layer cannot import commercial data, and a test fails the build if it ever does.
The site carries display advertising. Ad placements are decided by the page's data density, not by its subject, and no advertiser sees or influences the dataset.
The full account of how this site earns, and why no commercial arrangement can move a row up a table, is on the disclosure page.
Corrections
If a price here is wrong, it is a bug in a JSON file, and fixing it is a one-line change. Every page links to the vendor page the figure came from — if the two disagree, the vendor is right and this site is wrong. Report it and it will be corrected in the next deploy.