free-proxy-list.net won our test. On 9 September 2026 we fetched eleven free proxy lists and made real requests through each. Its addresses relayed 34.2 percent of the time in the first run that day. Thirty minutes later, on the same day, the figure was 38.3 percent. Nothing else came close. It is also the only list we tested that dates every single row. That per-row date is real, and it is smaller than it looks.
Two promises that sound alike
Read in September 2026, the site says its free list is "updated every 10 minutes". It also prints a Last Checked column, with values like "1 sec ago". Those are not the same claim.
The first one describes the page, and tells you how often the whole file gets rebuilt. The second one describes a single address, and says when that address last answered.
Almost every list makes the first promise. Very few make the second. free-proxy-list.net makes both, and that is most of why it wins.
The gap between the two matters, because these lists decay in an uneven way. A file rebuilt ten minutes ago can still be full of dead rows. A row checked one second ago was alive one second ago. Neither sentence says a thing about the next hour.
What eleven lists did on 9 September 2026
We took a random sample from each list and opened every address in it. The table is the first run, plus the run we repeated thirty minutes later on the same day.
| List | Addresses published | First run | Thirty minutes later |
|---|---|---|---|
| ProxyNova (decoded) | 26 of 37 rows | 53.8% | not re-run |
| free-proxy-list.net | 300 | 34.2% | 38.3% |
| ProxyScrape | 624 to 671 | 26.7% | 33.3% |
| Spys.me | 400 | 25.0% | 13.3% |
| monosans (GitHub) | 619 | 20.8% | 12.5% |
| iplocate (GitHub) | 646 to 718 | 18.3% | 16.7% |
| Geonode | 500 | not parsed | 7.5% |
| TheSpeedX (GitHub) | 3,024 | 7.5% | 5.0% |
| Proxifly (GitHub) | 2,385 | 2.5% | 4.2% |
| FreeProxy.World | 50 | not parsed | 2.0% |
| clarketm (GitHub) | 400 | 0.8% | 0.0% |
ProxyNova scored higher, but we could only decode 26 of its 37 rows that day. That sample is too small to rank against a list of 300. Two lists never loaded at all. OpenProxy.space returned HTTP 521 and proxy-list.download returned HTTP 502, both on 9 September 2026.
Size did not predict quality on the day we tested. TheSpeedX published 3,024 addresses that day and relayed 7.5 percent. free-proxy-list.net published 300 and relayed 34.2 percent. The shorter list did ten times the work.
FreeProxy.World shows the sharpest version of the problem. On 9 September 2026 its front page held 50 addresses. Thirty-nine of them belonged to Cloudflare, a content delivery network. Two more ended in zero, which names a network and not a machine. One of the fifty relayed that day, which is 2.0 percent.
What a timestamp cannot see
A check tells you about the past, and our own pool shows what the past is worth.
We run a free pool and keep its full record, which on 5 September 2026 held 763,409 addresses. By that date 691,179 of them had a recorded death. On the same day 9,451 were alive, 72,230 were in rotation, and we had run 115,078,758 checks.
Median life was 144.5 hours, about six days, on 5 September 2026. In that same record, 22.6 percent died inside the first hour. Another 39.0 percent lasted one to seven days, and only 17.0 percent got past a month.
Being alive is not the same as being dependable. The median live address passed 45.5 percent of its own checks on 5 September 2026. Of 9,452 live entries that day, 401 passed ten checks at 90 percent success.
Speed was the next problem on that date. Median latency was 1,050 ms on 5 September 2026. On that day 24.1 percent answered under 500 ms, and 28.1 percent took over three seconds.
Then comes the question of what your target already believes. On 5 September 2026, 89.3 percent of live entries with fraud data held a recent abuse flag. Their average fraud score was 72.5 on that date. None were residential, and 82.9 percent sat on datacenter address space that day.
One more figure explains why a fresh row is not a private row. On 5 September 2026, 116 sources fed our pool. The average live address showed up on 17.86 of them that day. 94.7 percent of them showed up on more than one.
So a dated row proves the address answered. It cannot promise the address will be there in an hour. It cannot say whether your target has already flagged it. It cannot tell you how many other people took the same row this morning.
For balance, our own pool checks constantly. On 5 September 2026 all 72,230 rotating entries had been checked within 24 hours. That day 14,762 of them had been checked in the previous five minutes. That is the same promise free-proxy-list.net makes, and it carries the same limits.
What to do with a dated row
A per-row date is useful if you treat it as a starting gun. Take the newest rows first, rather than the whole file. Re-check every address yourself, right before you use it. Expect the address to be shared. On 5 September 2026 the average live one sat on 17.86 sources. Retire an address the moment it fails twice, instead of retrying it. None of that needs a purchase, and all of it beats trusting the column.
Where that leaves free-proxy-list.net
The incumbent earned its rank on 9 September 2026. It published 300 addresses that day, and a third of our sample worked. Read in September 2026, its own page advertises "Total 6087 IPs in 2198 Subnets in 98 Countries". It sorts rows into elite, anonymous and transparent, and it dates each one. That is more than any other list we tested gave us.
What it does not print is the decay curve behind those rows. No list prints it, because measuring it costs far more than running the page.
Method and limits
We ran the test from one datacentre in Germany on 9 September 2026. For each list we fetched its own machine readable file. We parsed every address, then took a random sample of up to 120. We opened each address, spoke its protocol, and fetched a real page through it. The reply had to carry an exit address that was not ours. A port that merely accepts a connection does not count. We repeated the whole run thirty minutes later.
In the first run that day, 163 of 961 sampled entries relayed, which is 17.0 percent. The second run gave 158 of 1,131, which is 14.0 percent.
The limits are real. At 120 samples near a 17 percent rate, the error runs about 3.4 points. On 9 September 2026 the median swing between runs was 2.5 points, the largest 11.7. So the big swings are churn, not sampling noise. One country, one network, one afternoon. Your own numbers will differ. A list that failed to load for us may load for you.
Sources
- free-proxy-list.net, its own front page and its Last Checked column, read September 2026
- ProxyNova, ProxyScrape, Spys.me, Geonode and FreeProxy.World, lists fetched 9 September 2026
- GitHub lists from monosans, iplocate, TheSpeedX, Proxifly and clarketm, fetched 9 September 2026
- OpenProxy.space and proxy-list.download, both unreachable on 9 September 2026
- HProxy free proxy pool records, queried 5 September 2026
If you would rather not run any of this yourself, our paid pools start at $0.44/GB.