Free Proxy Lists for Web ScrapingWhat 640,600 Measured Proxies Reveal

If you searched for a free proxy list, you will find one in the next thirty seconds. There are dozens of sites publishing tens of thousands of IP and port pairs, updated by the minute, filterable by country and anonymity level, downloadable as a text file. The supply is real and it is genuinely free.

The harder question is what happens after you paste those IPs into a scraper. That part has actually been measured, twice, at scale, by academic researchers who published their methods. One study tracked more than 107,000 listed open proxies across 13 million requests. Another followed 640,600 of them across 30 months. Their findings are considerably more useful than another ranked list.

This post covers both halves: the free proxy lists that exist and what each one actually gives you, then what the measurements say about reliability, traffic integrity, and the real cost of running on them. It ends with the part most articles on this keyword skip, which is that "free proxy" describes two completely different products, and only one of them belongs anywhere near production.

Two different things are called a free proxy

The first category is the open proxy aggregator list. A publisher scans the internet for hosts that will forward traffic for anyone, checks whether they respond, and publishes the ones that do. The publisher does not own those machines and has no relationship with whoever does. You are borrowing a stranger's server, and neither you nor the publisher knows why that stranger is willing to let you.

The second category is the free tier or trial of a commercial provider. A company that operates a proxy network gives you a metered slice of it at no cost: a fixed number of requests, a bandwidth cap, or a time-limited trial. The IPs come from infrastructure the company sources, documents, and can be held accountable for.

Both get called "free proxies" and they behave nothing alike. Every measurement below describes the first category. Understanding that distinction is most of what separates a scraper that works from one that quietly collects garbage.

The free proxy lists, named

These are the aggregators that show up most often for this search, with what each one actually provides. Pool figures are the publishers' own claims at the time of writing and they move constantly, which is itself part of the story.

Read the fine print on those sites and something becomes clear: most of them tell you exactly what this is. ProxyScrape states that "the proxies we provide are public proxies, this means the proxy may be infected." Geonode warns that free public proxies "may log traffic, inject ads, or be honeypots" and adds, "never send sensitive data through free proxies." free-proxy-list.net describes its own inventory as "unstable and usually slow but very cheap" and notes the proxies do not work against major sites.

The publishers are not the problem here, and it is worth being precise about that. They index machines; they do not operate them. Everything measured below is a property of the anonymous operators whose servers end up on these lists.

What the measurements show: almost none of them answer

The headline number on a proxy list is not a count of working proxies. It is a count of entries.

In An Extensive Evaluation of the Internet's Open Proxies (Mani, Vaidya, Dworken and Sherr, ACSAC 2018), researchers pulled more than 107,000 listed open proxies from aggregator sites and pushed 13 million requests through them over 50 days. Their finding: more than 92% of open proxies that appear on aggregator sites are unresponsive.

Free Proxies Unmasked (Mehanna, Rudametkin, Laperdrix and Vastel, MADWeb 2024) went longer and wider, following 640,600 proxies from 11 providers across 30 months. Their availability figure: only 34.5% were active even once during the entire observation window.

Those two studies bracket the answer. Somewhere between a third and a twelfth of a published list responds at all, and responding once is not the same as being usable. Proxyway's hands-on check of 35 addresses from one popular list found 15 working initially and 11 still working half an hour later.

This is why free pools feel workable in a five-minute test and fall apart in a scheduled job. A handful of live proxies is enough to fetch one page by hand. A refresh cycle that needs sustained throughput spends most of its time timing out against addresses that stopped answering after you downloaded the list.

What the measurements show: some of them alter your traffic

Availability is the boring failure. The interesting one is that a measurable share of these hosts modify what passes through them.

The MADWeb 2024 study found 16,923 proxies manipulating content out of the 640,600 it tracked. It also scanned the proxy IPs themselves and catalogued 4,452 distinct vulnerabilities, including 1,755 that allow remote code execution and 2,036 that allow privilege escalation. Most telling for the question of who runs these things: 42,206 of them appeared to be running on MikroTik routers, which is to say consumer and small-business network equipment rather than anything resembling a proxy service.

The ACSAC 2018 study documented the specific behaviors. Researchers observed TLS man-in-the-middle attacks, HTML rewritten to inject cryptocurrency mining code, and binaries modified in transit to carry remote access trojans. As a control, the same team measured Tor exit relays and found no instances of TLS interception or content manipulation, which is a useful reminder that the problem is the incentive structure of anonymous open proxies specifically, not relayed traffic in general.

Why they are free

The word "free" implies someone is absorbing a cost. With aggregator proxies, four explanations cover nearly all of it, and only the first is innocent.

• Misconfiguration. The host was never meant to be an open proxy. Default settings on a router or a caching server left it forwarding traffic for the entire internet. The 42,206 MikroTik devices in the MADWeb data are this case, at scale, and the owners almost certainly have no idea.
• Harvesting. Running a proxy is a cheap way to collect a stream of other people's requests: hostnames, headers, cookies, session tokens, and anything sent without encryption. The traffic is the product.
• Injection. If content passes through unencrypted, it can be edited. The published findings include advertising, affiliate rewrites, and mining scripts, which monetize your bandwidth and your users' CPU.
• Observation. Some open proxies are honeypots, run by researchers, security vendors, or network operators who want to see what flows through them. Benign in intent, still an audience for your traffic.

None of these require anyone to be a sophisticated attacker. They require a machine that forwards packets and a list that advertises it.

The arithmetic of a free pool

The case for a free list is that it costs nothing. That holds only if your time is worth nothing and failed requests are free, and neither is true past a prototype.

Work the model with your own numbers. If a job needs S successful fetches and the pool converts at rate p, you issue roughly S / p attempts. At the measured survival rates above, a p between 0.3 and 0.4 is optimistic for a list you downloaded an hour ago, so a 50,000 page job becomes 125,000 to 165,000 attempts, most of which are timeouts rather than fast rejections. Timeouts are the expensive kind of failure because each one holds a worker for seconds.

Then add the parts that do not show up in a request counter:

• Re-downloading and re-validating the list, because entries expire while you use them.
• Building and maintaining health checks, retry logic, and dead-proxy eviction. This is a small distributed system, and you now own it.
• Re-running batches that completed with silent gaps, which is the failure mode that costs the most because you find it downstream in the data.
• Verifying that nothing in the collected corpus was modified in transit.

An engineer spending two days a month on that has spent more than a paid plan costs, and has produced no product. Our rubric for scoring proxy providers walks through the same trade in more detail.

When a free proxy list is the right tool

There is a real answer here, and pretending otherwise would be dishonest. Aggregator lists are genuinely useful when the stakes are zero.

• Learning the mechanics. Configuring a proxy in curl, in requests, or in your language's HTTP client, and watching your apparent IP change, is a worthwhile exercise and free proxies are fine for it.
• One throwaway fetch of a public page. No credentials, no session, nothing you care about.
• Checking whether public content varies by country, as a rough signal before committing to real geotargeting.
• Testing that your own code paths handle proxy failure, where an unreliable pool is arguably the ideal test fixture.

The line to hold: never send anything through an unknown proxy that you would mind a stranger keeping. That means no logins, no API keys, no cookies, no personal data, and nothing whose integrity you plan to trust later. Our post on whether proxies are safe covers the sourcing questions to ask any provider, and the rotating residential landscape covers where free pools sit relative to paid ones.

The free tier that actually works

Back to the distinction this post opened with. If what you want is to scrape without paying yet, the answer is not a list of strangers' routers; it is a metered slice of infrastructure that has an owner. Several exist. Webshare gives 10 static datacenter IPs and 1 GB a month. Oxylabs offers a US-only free tier with a 5 GB monthly cap.

The Crawlbase Smart AI Proxy gives you 5,000 credits with no credit card, against a pool of 140M residential IPs with geotargeting across 45+ countries, an average success rate of 99%, and unlimited bandwidth. Blocks and CAPTCHAs are handled inside the endpoint rather than in your retry loop.

The operational difference matters more than the pool size. There is one hostname and one port. No list to download, validate, rotate, or evict, because rotation happens upstream and every request arrives from a fresh IP.

curl -x 'https://YOUR_TOKEN:@smartproxy.crawlbase.com:8013' \
     -k 'https://httpbin.org/ip'

Your token is the username and the password is left blank. Port 8013 is the HTTPS proxy and is the recommended one; 8012 speaks plain HTTP to the proxy for clients that require it. Both handle HTTP and HTTPS targets. The dictionary key names the target scheme, and each entry uses the scheme its port expects: the HTTP proxy answers on 8012, the HTTPS proxy on 8013. The same thing from Python:

import requests

proxies = {
    'http':  'http://YOUR_TOKEN:@smartproxy.crawlbase.com:8012',
    'https': 'https://YOUR_TOKEN:@smartproxy.crawlbase.com:8013',
}

res = requests.get('https://httpbin.org/ip', proxies=proxies, verify=False)
print(res.text)

Per-request geotargeting, device type, and sticky sessions are set with request headers rather than by swapping endpoints, and if you want the rotation without the AI layer that is a configuration choice, covered in standard rotation versus the AI layer.

Choosing between the options

If you are still deciding what kind of IP a job needs before picking a vendor, datacenter versus residential and matching the proxy type to the target are the two to read next. For the mechanics of rotation itself, per-request versus sticky sessions has the code.

The short version

Free proxy lists are real, plentiful, and mostly dead. Two peer-reviewed studies put availability between 8% and 34.5%, and the publishers themselves describe the inventory as public machines that may be infected. Thousands of those hosts have been observed modifying traffic, tens of thousands of them are somebody's misconfigured router, and the labor of keeping a pool of them alive costs more than a plan.

If the requirement is genuinely "no budget yet," take a metered free tier from an operator with a name on it. You get a working IP on the first request instead of a queue of timeouts, and nobody is reading your traffic as the price of admission.

Frequently asked questions

Are free proxies safe for web scraping?

Not for anything you care about. Peer-reviewed measurement of 640,600 free proxies found 16,923 modifying the content passing through them, and an earlier study of 107,000 documented TLS interception, injected cryptocurrency miners, and binaries rewritten in transit to carry remote access trojans. Several list publishers say so themselves; ProxyScrape's own page notes that its proxies "may be infected." Use them for learning and throwaway fetches of public pages, never for logins, API keys, cookies, personal data, or any corpus whose integrity you intend to trust.

Why are free proxies free?

Four reasons cover most of them. Many are simply misconfigured: one study found 42,206 apparently running on MikroTik routers whose owners almost certainly did not intend to operate a public proxy. Others exist to harvest the traffic that flows through them, to inject advertising or mining code into unencrypted pages, or to observe traffic as honeypots. Somebody is paying for the bandwidth, and the traffic is usually the payment.

How many proxies from a free list actually work?

Far fewer than the headline count. The ACSAC 2018 study found more than 92% of aggregator-listed open proxies unresponsive. The MADWeb 2024 study found only 34.5% of 640,600 proxies active even once across 30 months. Attrition also continues while you work: an independent hands-on check of 35 addresses from one list found 15 alive initially and 11 still alive 30 minutes later. A published pool of 20,000 is a table of 20,000 rows, not 20,000 usable IPs.

What is the difference between a free proxy list and a free proxy trial?

Accountability. A list indexes machines run by strangers who have no agreement with you or with the publisher. A trial or free tier is a metered piece of a network that a company operates, documents, and can be held responsible for. The words are similar and the products are not comparable; every reliability and integrity finding in this post applies to the first kind.

Can I use a free proxy list for a production scraper?

You can, and the cost surfaces elsewhere. With a survival rate around a third, every 50,000 successful fetches needs roughly 150,000 attempts, mostly slow timeouts, plus list re-validation, health checking, dead-IP eviction, and reruns for batches that finished with silent gaps. That is an unpaid distributed systems project sitting underneath your actual work, and it usually costs more in engineering time than a plan costs in money.

Does using HTTPS make a free proxy safe?

It helps and it is not sufficient. Encryption stops a passive proxy from reading or editing the page body, which neutralizes most of the injection behavior researchers documented. The operator still learns every hostname you request and when, can stall or drop connections selectively, and can offer its own certificate to any client that skips verification. HTTPS reduces the blast radius; it does not turn an anonymous host into a trustworthy one.

VP

Viktor Petrov

Data Science Architect · Crawlbase

Data Science Architect at Crawlbase. Works on the models that keep large-scale crawling reliable against modern bot defenses.