About
One person, one server, one script
OpenWebData compares web scraping proxies by testing them. Not by reading the sales pages, not by asking the providers. The same requests go to the same nine sites from the same machine on the same day, and whatever comes back is what gets published, including the runs where the provider we earn the most from loses.
Latest run
- Tested
- 10 Sep 2026
- Targets
- 9real sites
- Providers
- 3+ datacenter baseline
- Baseline pass rate
- 15 / 45no proxy
Who does this
Sebastian Rantala
I run this site and I run the tests. I have built and maintained websites and server infrastructure since 2012 through my company, Bastinet, a registered Finnish business (ID 2492063-7). The test server, the scripts and the proxy subscriptions are paid for out of that business.
I am not a proxy reseller and I do not work for any provider. I started measuring because I needed the answer myself: I scrape public pages for client projects, the provider comparisons I could find were written by the providers or by people who had clearly not run a request, and nobody published a number I could check.
If a result here looks wrong to you, the raw output of every run is published and the method is written down. Tell me what I got wrong and I will rerun it: sebastian@bastinet.fi.
Method
How the numbers are produced
The same script, the same day, the same server. Anything else would be an opinion.
The test machine is a virtual server at Hetzner (AS24940) in Helsinki. Requests are made either by curl with current desktop Chrome headers and no JavaScript, or by a headless Chromium through the same proxy, and both runs are published separately because they give different answers. A response counts as a pass only when it contains a marker that exists on the real page and nowhere else. Timeouts, connection resets, redirects to a block page and full-size pages with empty result containers are failures, and the reason is recorded. Exit addresses are looked up per request, so rotation and the owning network are measured facts rather than claims. Prices are list prices read from the provider’s own pricing page on the test date.
Money
How this site is paid for, in plain terms
If you buy a subscription through a pricing link here, the provider pays a commission. That is the only revenue this site has. Partner links are marked as sponsored in the page source, and the disclosure sits directly above the results table, not in the footer.
A position, a score, or an early look. Providers do not see results before you do and cannot pay to appear. The ranking is the pass rate, computed from the run, and the provider that pays the most commission is not the one at the top of every table.
The proxy subscriptions used in testing are bought at list price like anyone else’s, out of the business. No provider has supplied a free account, and if one ever does, that will be written on the page where its numbers appear.
Because a comparison you cannot check is a review, and reviews are cheap. Every run is published as JSON with each request, its verdict and the reason the site gave. Exit IP addresses are masked to /24, because they belong to real households.
Limits
What this site does not tell you
Every measurement has an edge. These are the ones worth knowing before you act on a number here.
- Nine sites is not the internet. The targets are picked because they are commonly scraped and defended by different systems. If you scrape something else, use the target matrix as a guide to how a defence behaves, not as a promise about your site.
- One day is one day. Sites change their defences continuously, and a provider’s pool changes with it. The date on every table is the date of the numbers, and the test is rerun monthly rather than written once.
- US residential, rotating. That is the product tested. Mobile, ISP and datacenter products from the same providers behave differently and are not covered by these numbers yet.
- No login, no cart, no scale test. Every request is anonymous, and the test is a measurement, not a load test. How a provider behaves over a million requests a day is a different question and needs a different test.
- Pool sizes are the providers’ own claims. They are printed in the plan table because buyers ask, and they are the one number on this site that is not measured.
Contact
Corrections, questions, provider replies
A provider who thinks a run was unfair is welcome to say so, and the reply will be published next to the result.
OpenWebData is operated by Bastinet, a registered Finnish sole proprietorship (business ID 2492063-7), Finland. Editorial policy: results are published as measured, including results that cost us commission. Corrections are made in the open, with the date of the change.