Best Free Proxy List in 2026: We Tested 2,911 Proxies

Across 2,911 proxies pulled from the free lists of seven commercial proxy providers and probed three times each, databay.com ranked first with 63.9% of its list alive, the fastest median latency (1067 ms), and the highest stability (73.4% of its round-1 survivors stayed alive through all three rounds). ProxyScrape came second at 36.7% alive. Every other list was below 21% alive, and the bottom three were under 8%.

Two disclosures before any of this counts. We run databay.com, and our own list is one of the seven tested. That is exactly why the entire benchmark (harness code, the frozen proxy snapshots, and the raw per-round results) is open source on GitHub. Don't take our word for it; re-run it.

And the result that matters more than the ranking: even the winning list left 36.1% of its entries dead at test time. Most lists were 80–98% dead. Free proxies are for experiments, not for anything you depend on.

Nearly every commercial proxy provider publishes a free proxy list as lead generation, ours included. Every one of them claims its list is "fresh" and "verified." None of them publish head-to-head numbers. A free proxy list is a snapshot of things that are dying, and the numbers are unflattering for the whole category, ours included.

So we measured it: take the freshest proxies each provider offers, freeze that snapshot, probe every proxy the same way from the same machine at the same time, and publish everything, including the parts that make free proxies as a category look bad. The practical questions we wanted answered: how much of a free list works right now? Can it tunnel HTTPS? Can it reach the sites people actually use it for, like Google and Amazon? Does it hide your IP? And does it still work ninety seconds later?

On June 11, 2026 (02:29–02:53 UTC) we loaded up to 500 of the freshest proxies from each of seven lists, 2,911 proxies in total: databay.com 497, ProxyScrape 500, Geonode 500, FloppyData 314, RoundProxies 100, NodeMaven 500, Proxiware 500. Databay's sample was its most recently checked proxies (sorted by lastChecked); the other API-based lists were also sampled freshest-first, so every list is shown at its best. List sizes differ, so all comparisons in this post use rates, not raw counts.

Every proxy was probed three times, 90 seconds apart, from a single host in Thailand, using the protocol (HTTP, HTTPS, SOCKS4, SOCKS5) each list itself declared. Per round, each proxy was checked for:

• Alive: it relayed a real request to ip-api.com (HTTP) or api.ipify.org (HTTPS).
• Latency: the fastest successful round trip, in milliseconds.
• HTTPS: it completed an HTTPS CONNECT tunnel.
• Google pass: https://www.google.com/generate_204 returned HTTP 204 through the proxy.
• Amazon pass: https://www.amazon.com/robots.txt returned HTTP 200. Amazon blocks most datacenter and proxy IPs, which makes this the hardest real-world test in the suite.
• IP hidden: the exit IP the destination saw differed from our real IP.
• Anonymity: elite / anonymous / transparent, judged best-effort from header reflection at httpbin.org.
• Stability: of the proxies alive in round 1, how many stayed alive in all three rounds.

The composite score blends these into a 0–100 number: 30% alive + 20% Google + 15% Amazon + 10% HTTPS + 10% IP-hidden + 15% latency (latency normalized so 0 ms scores 1.0 and a median of 5000 ms or more scores 0). Because the test re-probes a fixed IP:port snapshot, the whole run is reproducible: the snapshots and the raw per-round JSONL are committed to the benchmark repo.

The headline pattern isn't that databay won; it's the shape of the curve. Exactly two lists had a usable share of working proxies: databay.com at 63.9% alive and ProxyScrape at 36.7%. Then the floor drops: FloppyData 20.3%, Proxiware 12.9%, NodeMaven 7.7%, RoundProxies 4.7%, Geonode 2.5%. A random pick from any of the bottom three lists had, at best, a one-in-twelve chance of relaying a single request at capture time.

The harder checks separate the field further. Roughly one in three databay proxies (32.4%) could fetch Google's generate_204 endpoint, and one in six (16.8%) survived Amazon's notoriously proxy-hostile front door. Outside the top two lists, Google reachability never exceeded 11.9% and Amazon never exceeded 3.8%.

Geography matters too. databay's live exits geolocated across the US (158), China (76), Japan (45), France (44), and Hong Kong (36); ProxyScrape's skewed heavily American (US 200, then France at 26); NodeMaven's working proxies were mostly British (GB 62, US 36). If you need geographic spread from a free list, the exit-country distribution of the proxies that actually work is worth checking before the advertised one.

Any list can look acceptable for one probe. The question is what happens when you come back. We probed the same frozen snapshot three times, 90 seconds apart.

databay's alive rate held flat across the rounds: 64.4%, 63.4%, 64.0%. Of its 320 round-1 survivors, 235 (73.4%) stayed alive through every round. ProxyScrape held 46.1% (88 of 191). FloppyData, with a much smaller working set, held a respectable 64.1% (41 of 64).

NodeMaven is the cautionary tale. It opened at 16.2% alive, then collapsed to 3.0% by the second round and 4.0% by the third. Only 13 of its 81 round-1 survivors (16.0%) held through the run. Whatever produces that pattern, the practical effect is that the proxies you fetch are dying before you finish pasting them into a script.

RoundProxies (3, then 4, then 7 alive out of 100) and Geonode (14, 10, 13 out of 500) were close to statistical noise: at a 2.5–4.7% alive rate, the error bars matter more than the rank.

Mostly no, and where they do overlap, it shows in the scores. We computed Jaccard similarity between the seven IP:port snapshots. The highest overlap was Geonode ∩ Proxiware at 0.322 (243 shared endpoints), followed by Geonode ∩ RoundProxies at 0.200 (100 shared). databay ∩ ProxyScrape, the top two performers, overlapped at just 0.126 (106 endpoints), and most pairs sat below 0.10.

The reading we'd defend: there is a common pool of endlessly re-scraped public proxies, and the lists that lean on it hardest were also the deadest at test time. databay's working set barely overlaps anyone else's, which is consistent with it coming from our own continuous scraping-and-pruning pipeline rather than a re-export of the common pool. The pipeline is documented in our free proxy list methodology post.

"Alive" is a low bar, so the benchmark also measured what the destination could see. The robust signal first: for every list, every proxy that was alive also hid our probe IP from the destination, so the IP-hidden rate equals the alive rate across the board.

The finer elite / anonymous / transparent split is best-effort, because it depends on a header-reflection judge (httpbin.org) answering through proxies that are often barely functional. Among databay's 953 alive checks across the three rounds, 604 judged elite, 129 transparent, and 220 unknown; ProxyScrape's split was 282 elite / 21 transparent / 247 unknown. databay shows the largest absolute transparent count simply because it had by far the most live proxies; as a share, 13.5% of its alive checks leaked our real IP in a request header. A transparent proxy hides nothing that matters: the destination sees both the proxy and you.

Which brings us to the warning that belongs in every article about free proxies, including ours: never route logins, payments, or any sensitive traffic through a free proxy. Every exit in every one of these lists, databay's included, is operated by someone you don't know, who can observe any traffic that isn't end-to-end encrypted and can tamper with plain HTTP at will. Treat every free proxy as hostile infrastructure that happens to be useful. The full risk rundown is in Are Free Proxies Safe?

The numbers above are only as good as the test design, so here is everything that should make you cautious:

• We ran the test, and we're in it. databay.com built the benchmark, and databay's list is one of the seven measured. The counterweight is full reproducibility: harness code, the exact frozen IP:port snapshots, raw per-round JSONL, and the report generator are all in the public repo.
• Snapshot in time. Free proxies churn by the minute. These numbers describe June 11, 2026, 02:29–02:53 UTC. A re-run on live lists will differ; re-probing our frozen snapshot reproduces the analysis, not the liveness.
• Freshest-first sampling, both ways. databay was sampled by most recent lastChecked; the other API lists were also taken freshest-first. Every list is shown at its best.
• Unequal list sizes. RoundProxies published only 100 proxies and FloppyData 314 at capture time; the rest were 497–500. We compare rates, and small samples carry wide error bars; RoundProxies' 33.3% stability is literally 1 proxy out of 3.
• Single vantage point. Everything was probed from one host in Thailand. Reachability varies with geography; a proxy that timed out from Bangkok might answer from Frankfurt.
• Anonymity is best-effort. The elite/transparent split depends on httpbin.org responding through each proxy; many returned "unknown." The IP-hidden rate is the more trustworthy number.
• Fixed probe parameters. 60 concurrent probes, 4 s connect / 6 s read timeouts. Looser timeouts would lift every list's alive rate together; the ranking is what's robust.

Sources tested, as published on capture day: databay.com's free proxy list API, ProxyScrape, Geonode, FloppyData, RoundProxies, NodeMaven, and Proxiware. Anything not in this post (per-proxy results, exit-country tables, latency percentiles) is in the full dataset in the repo.

If you need a free proxy list in 2026, databay.com's was measurably the freshest of the seven we tested, and it still handed you a list where roughly one proxy in three was already dead. That is the ceiling for free. The median free list in this test was more than 87% dead, slow when it did answer, and frequently unable to reach the sites people actually want.

So use free proxies for what they're good at: throwaway experiments. Checking a geo-block, a one-off curl, learning how proxies work. The moment a task has consequences (sustained scraping, anything tied to an account, anything where a 64%-at-best success rate compounds into failure), the economics flip, and a paid pool with an accountable operator, like residential proxies, can cost less than the failed retries.

Grab the free list, re-run the benchmark, check our work. And when you outgrow free, you know where we are.