The free proxy industry, explained

Nobody publishes thousands of addresses for free out of generosity. Our census of the sites that rank or ranked for “free proxy list” holds 69 of them (6 September 2026). 11 are run by companies that sell proxies, and 10 of those are alive. 52 are independent, and 21 of those are dead, parked, for sale or absorbed into another site. The sellers hold the demand: one of them draws about 165,000 searches a month for its name, the biggest independent about 22,000. The pattern is the industry in one sentence: the lists that survive are the ones with something to sell.

The model is printed on the pages. We read each list as text from our own server on 16 September 2026 and copied the sentence that sells.

Three business models cover the whole industry. A paid provider publishes a free table so that a visitor who finds it useless buys the paid pool that sits beside it; the free rows are the argument for the price. An independent site sells the ads around the table, a members’ version of it, or a VPN deal. A GitHub repository sells nothing and is fed by a script that copies the other two. The oldest names show where the model ends: Proxy4Free, registered in 2002, and ProxyElite, online since 2007, carry no list at all any more and sell proxies under the old name.

Each list was fetched once from our server on 16 September 2026, between 13:45 and 16:00 UTC, and parsed into addresses. Every address was then looked up in our own records, which hold a check every ten minutes for the hot pool and a daily re-check for the archive, and asked once more to relay a request to our own echo page. Three columns follow: the share of each list that never worked once in our records, the share our verifier found alive at its last check, and the share that relayed the fresh request. The fresh request makes one attempt, with the listed protocol, inside ten seconds. The verifier is more patient, so its number is higher. The stricter number is the one a reader experiences.

The two largest claims are the two largest gaps. One provider’s page says the list “typically holds 20,000+ live proxies across 180+ countries at any time”; its own API served 1,895 distinct addresses that hour, and 519 of them relayed our request. Another provider’s page title says “5,500 Live IPs, Updated Every Minute” and the page prints “5,508 Proxies Online”; its own API, asked the same hour, answered a total of 1,983, and 55 of the 1,918 entries relayed. The biggest count of all, “Found 45226 proxies”, belongs to a site whose first thousand rows include 957 addresses that have never worked once in 252,689,860 checks.

The most copied list on the web, the 300-row table stamped “Just Checked” with rows marked “14 secs ago”, does better than its imitators and still not well: 86 of 298 rows relayed our request, and 98 of them had never relayed anything for us. In the thirty days before 16 September 2026 that site published 8,427 addresses we had not seen before, and 4,749 of them (56.4 percent) never answered a check.

One day is a snapshot. We have also watched the largest lists every day since 13 May 2026, recorded every address they published, and checked each address every few minutes for as long as it stayed reachable, then daily after it went quiet. That record answers the question a snapshot cannot: of everything a list has published in four months, how much ever worked, and how much still does.

Three of every four addresses the largest list published this summer worked at some point, and 1.33 percent of them work today: 292,710 addresses, 119,257,500 checks, one in seven successful. The 300-row list is the opposite shape: barely half of what it published ever worked, because it recycles addresses that have been dead for months, but the ones that work last longest, a median 1,681 hours. The curated list of a paid provider published the fewest addresses, 10,279, and keeps the highest share of them alive, 11 percent, because it republishes what everyone else already found working: all but two of its 10,279 addresses also appear on other lists.

Almost every free proxy is a server somebody left open by mistake: a forgotten test box, a misconfigured router, a cloud instance with a proxy daemon and no password. The lists do not find them; scanners do, and the lists copy each other. We have watched 116 public lists since 13 May 2026 (103 of them still publish), and the overlap between them explains the industry better than any single list does.

An address that only one list ever carried is alive in 0.27 percent of cases. An address that fifty lists carry is alive in about 10 percent. The working addresses are the same few thousand everywhere, republished until they die, and the long tail of a big list is the noise its scraper collected from other scrapers. That is why the biggest lists have the worst yield: across the 116 lists we read, the best has 16.8 percent of its addresses alive and the worst 0.54 percent.

The other source is people’s devices. The residential proxy business, studied by Mi and colleagues at the IEEE Symposium on Security and Privacy in 2019, showed 6 million residential addresses across more than 230 countries, with 237,029 of the hosts identified as IoT devices, while the providers said every host had joined willingly. The indictment of the 911 S5 operator in the Eastern District of Texas (2023) records that the network was built by distributing free VPN applications named ProxyGate, MaskVPN, DewVPN and Shine VPN, “with the intent to infect residential computers worldwide”. A free VPN that turns the phone into an exit is the answer to why a free proxy is free.

Of the 121,503 addresses in our records that worked at least once and then died, the median lived 251 hours between the first sighting and the last success, about ten days. A quarter were gone within 32.8 hours. The upper quarter lasted more than 1,066 hours, and those are the addresses every list keeps: the 1,986 live addresses older than ninety days are the backbone of the whole free pool.

A cohort tells the same story day by day. Of the 5,027 addresses first published between 13 to 17 August 2026, 1,923 answered on their first day. This is how many of those still answered on the days that followed.

Alive is not the same as reliable. The 5,447 addresses that were alive on 16 September 2026 and have thirty days of history answered a median 33.4 percent of their probes in those thirty days; the middle half sits between 18.1 and 55.6 percent. Only 369 of them reached ninety percent, and 3,872 answered less than half the time. A free proxy that works right now will, on the median, fail two of the next three requests. The median response time of the live pool at its last successful check was 7.3 seconds.

On 16 September 2026 the verifier had 5,976 addresses alive at their last check: 3,205 HTTP, 1,520 that also relay HTTPS, 680 SOCKS4 and 1,388 SOCKS5. By anonymity, 3,861 presented as elite, 1,515 as anonymous and 603 as transparent, which means the destination sees the visitor’s real address in a header.

Elite describes headers, not reputation. 3,501 of the 5,979 live addresses sit on datacenter address space, and so do 2,383 of the 3,861 elite ones (61.7%). Of the 5,980 live addresses with fraud intelligence, 5,776 (96.6%) are flagged for recent abuse, with a mean fraud score of 71 out of 100. A site that checks reputation before it serves a page has already seen these addresses.

Countries the live pool exits from

United States

3,149

Indonesia

672

China

174

Russia

135

Germany

110

India

110

Brazil

106

Bangladesh

100

The pool is also young and old at once. The median live address was first published 20.9 days ago; 983 appeared within the last week and 1,986 have been on the lists for more than ninety days.

Our numbers are one engine’s view of one summer. Four measurements by people with no proxies to sell, made between 2015 and 2019, found the same population and the same shape.

The malicious share matters more than the dead share. Perino and colleagues found that the proxies which injected ads or intercepted TLS were also the fastest, because somebody was paying for them. Tsirantonakis and colleagues attributed the injections: ads, tracking code, and redirects to malware. Haschek found that four in five working proxies refused HTTPS in 2015, which kept the traffic readable. The rule that follows is short: nothing that requires a password goes through an address from a list.

Until 6 September 2026 our verifier counted a proxy as alive when it answered with any HTTP 200, whatever the body. A server that returns its own page for every request passed that test without relaying anything, which is exactly what the addresses in the decoy table do. When the judge started to require our own echo body, the pool’s daily count of addresses alive at least once fell from 57,256 on 6 September to 24,514 on 7 September. Every figure on this page is from the fixed judge, and the survival chart says which of its days are not. Our earlier studies on this site carry the older, larger numbers with their own dates.

The questions people bring to this page are the same ones they post on forums and send to our support: is any free proxy safe, why does a list with ten thousand rows give almost nothing that connects, why did a free address work when a paid one did not. The answers follow from the measurements above.

1. Ignore the headline count. Every list prints one, and on 16 September 2026 the gap between the count and the addresses that relayed a request ran from three to one up to sixty to one. Judge a list by the share that answers a checker today, and expect that share to be between one and forty percent.
2. Look for the last-checked time on every row. A list that shows when each address was last verified can be audited; a list that says “updated every minute” without a per-row time cannot. Even a per-row time is a claim to be checked: one list stamps rows “14 secs ago” that never relayed for us.
3. Verify before you use, every time. Run the addresses through a checker that asks the proxy to relay a real request and reads the answer, not one that counts any response. Our proxy checker does that for a pasted list, and this guide explains what the grades mean.
4. Watch for addresses that are not proxies. Rows in 104.21, 172.67 or 162.159 on port 80 are a content delivery network answering with its own error page. A checker that marks them alive is the checker that is wrong.
5. Never send a login or a payment through one. The operator is unknown, the traffic is readable to it, and one working proxy in ten has been found to inject or intercept. A free proxy is for seeing how a page looks from another country, and for nothing that carries a password.
6. Plan for two failures in three. The live pool answers a median third of its probes over thirty days. Pull ten times the addresses you need, rotate on failure, and refresh the set every few hours, because the median address is gone in ten days.

The free address that worked when a paid one did not is usually a sign that the paid one was configured wrongly, not that the free one is better; a paid endpoint fails on a wrong password or an allowlist, a free one fails at random. When free proxies are fine draws the line by task, and are free proxies safe covers the risks one by one.

The measurements set the limit. A free address is shared with everyone who found the same list, 97% of the live pool is already flagged for abuse, a session on a free address cannot be held because the address answers a third of the time, and nothing about its uptime is promised by anyone. Our free proxy list is the output of the verifier described on this page, re-checked every few minutes with the last-checked time on every row, and it is the right tool for a look at a page from another country. For a login, a purchase, a scrape that has to finish, or anything a company depends on, the same verifier watches our paid pool, and that is the product this site sells.

Each name below has its own page: what the list is, what its operator sells, how many of its entries were alive when we checked, and what to use when it stops refreshing. The lists are the subjects of those pages, not links.

The studies this page grew out of: 580,000 free proxies checked 110 million times (August 2026), 110 million checks on free proxies (July 2026), the biggest free proxy lists, ranked (July 2026) and what an open proxy is. Any single address can be looked up in the IP lookup tool.

We have watched 116 public lists every day since 13 May 2026 and recorded every address they published, whether it ever answered or not, together with the lists it appeared on. The verifier asks each address to relay a request to our own echo page and counts it alive only when the echo body comes back; the hot pool is probed every ten minutes, the archive daily, and every probe is kept in a per-day record, which is what makes the survival and uptime figures possible. The pool figures are read-only aggregates of that database on 16 September 2026. The list audit fetched each list once from our server in Germany between 13:45 and 16:00 UTC the same day, joined every entry to the records, and asked it once more to relay, with the listed protocol and a ten-second limit. Paginated lists were sampled from their first pages, and the sample sizes are printed in the table. Two lists could not be read in full because they challenge non-browser requests, and one renders its rows with a script; those three are quoted, not audited in full. Two aggregate reads of a database that is being written to every second disagree by a few rows, which is why some totals on this page differ by a handful.

Everything changes fast here. List sizes, headline counts and prices are quoted with the day they were read, and this page is refreshed against the same database and the same lists every quarter, next by 16 December 2026.

• Akshaya Mani, Tavish Vaidya, David Dworken and Micah Sherr, “An Extensive Evaluation of the Internet’s Open Proxies”, Annual Computer Security Applications Conference, December 2018.
• Diego Perino, Matteo Varvello and Claudio Soriente, “ProxyTorrent: Untangling the Free HTTP(S) Proxy Ecosystem”, The Web Conference, April 2018.
• Giorgos Tsirantonakis, Panagiotis Ilia, Sotiris Ioannidis, Elias Athanasopoulos and Michalis Polychronakis, “A Large-scale Analysis of Content Modification by Open HTTP Proxies”, Network and Distributed System Security Symposium, February 2018.
• Christian Haschek, “Analyzing 443 free proxies”, 21 June 2015.
• Xianghang Mi and others, “Resident Evil: Understanding Residential IP Proxy as a Dark Service”, IEEE Symposium on Security and Privacy, May 2019.
• United States v. YunHe Wang, indictment, United States District Court for the Eastern District of Texas, 2023; press release of the Department of Justice, 29 May 2024.
• The list pages of ProxyScrape, Geonode, Webshare, advanced.name, free-proxy-list.net, FreeProxy.World, hide.mn, spys.me, ProxyDB, proxy-list.org, Proxy4Free and ProxyElite, and the GitHub repositories of proxifly, TheSpeedX, monosans, iplocate and clarketm, read as text from our server on 16 September 2026.
• HProxy verifier and collector, pool database, read-only aggregates of 16 September 2026; the list audit of the same day; the verifier change record of 6 September 2026. Queries and raw results: docs/research-2026-09-16-free-proxy-industry in our repository.