Buying guides

Proxy pricing is unusually hard to compare because the units differ between categories and the definitions differ between providers. Two plans with the same headline price can cost very different amounts for the same job.

Match the billing model to the shape of the work

Per-address billing makes heavy use of a few identities cheap. Per-gigabyte billing makes light use of many identities cheap. Getting this backwards is the most expensive single mistake available.

• Many pages, each fetched once, spread widely → per gigabyte, so residential or mobile.
• The same identity used repeatedly over a long period → per address, so ISP or datacenter.

The category comparison works through the decision in more detail, including the cases where the cheaper option is simply correct.

Find out what is billable before you commit

Metering normally counts bytes crossing the proxy, not bytes you successfully used. Bandwidth billing sets out what that usually includes: response headers, failed and blocked requests, challenge pages, both hops of a redirect, every retry, and every image and font a rendered page pulls in.

Ask the provider directly whether failed requests are metered, and get the answer in writing. The definition varies, and it is the difference between a predictable invoice and a surprising one.

Which advertised numbers to discount

Two figures dominate proxy marketing and neither is comparable between vendors without its method.

Pool size has no standard definition. A total may count every address observed over a long window, most of which are no longer reachable. The useful questions are how many are available concurrently, how many in the country you need, and across how many distinct networks — because addresses concentrated on a few networks can be removed by one classification decision.

Success rate means nothing without five facts: sample size, target, date, definition of success and concurrency. A rate that counts any HTTP 200 as success is systematically flattering, because challenge pages and empty results return 200 too.

Test before you scale

Measure a representative sample of your own targets, not the pages you hope to collect, and include the failures and retries because those are billed as well. A trial that only measures the happy path will understate what the work costs.

We publish no provider pricing or performance tables here. Reprinting a vendor's figure as though it were a measurement would be worse than printing nothing, and no independent party has counted the pools.