ProxyScrape is one of the better free proxy lists we measured. On 9 September 2026 we sampled it twice, thirty minutes apart. In the first run 26.7 percent of the sample relayed real traffic. In the second run 33.3 percent did. Only free-proxy-list.net and ProxyNova did better that day. So the list works. Two thirds of it still does not, and that gap is what your script has to survive.
What a one in three list costs your loop
Treat the file as a queue and the arithmetic gets simple. At the 33.3 percent we measured on 9 September 2026, one entry in three answers. Ten working sessions cost about thirty dials, and about twenty of those fail. At the weaker 26.7 percent, the same ten sessions cost about thirty seven dials.
The dials are not the expensive part. A dead proxy rarely refuses fast, and it hangs until your client gives up. With a five second connect timeout, twenty failures cost you under two minutes. Push that timeout to thirty seconds and the same failures cost ten minutes. The cure is a short timeout and more dials at once, not patience.
Size is not supply
The biggest files scored worst on 9 September 2026. TheSpeedX published 3,024 entries and returned 7.5 percent. Proxifly published 2,385 and returned 2.5 percent. ProxyScrape published between 624 and 671 and returned 26.7 percent.
Multiply the count by the rate and the picture shifts again. ProxyScrape works out to roughly 170 to 220 live entries. TheSpeedX works out to roughly 150 to 230. The two sit close in raw supply, which surprised us. Both of those are estimates from a sample, not a full count.
The real gap is the price you pay per entry. ProxyScrape costs you about three dials for one working proxy. TheSpeedX costs you between fourteen and twenty. That is the number to watch, not the size of the file.
ProxyNova led on rate that day, at 53.8 percent. It published only 37 rows, and 26 of them decoded. A high rate on 26 usable rows is a very small pond.
The full board, both runs
Every figure in this table comes from one audit on 9 September 2026. It ran from a single data center in Germany.
| List | Entries 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% |
Across all eleven lists, the first run relayed 163 of 961 sampled entries, or 17.0 percent. The second run relayed 158 of 1,131, which is 14.0 percent. Two more sources never answered at all that day. OpenProxy.space returned HTTP 521 and proxy-list.download returned HTTP 502.
Why the rate moved 6.6 points in half an hour
ProxyScrape went up between the two runs, from 26.7 to 33.3 percent on 9 September 2026. Spys.me went the other way, from 25.0 down to 13.3 percent. Across every list that day the median swing was 2.5 points. The largest was 11.7 points.
A sample of 120 carries about 3.4 points of error near a 17 percent rate. So the small moves are noise, and the big ones are real churn. A list you pulled an hour ago is already a different list.
What the churn looks like from inside
We run a free proxy pool of our own, so we can watch the decay directly. On 5 September 2026 it had seen 763,409 entries since it opened. 691,179 of those had a recorded death. 72,230 were in rotation and 9,451 were alive. The pool had run 115,078,758 checks by that date.
Life spans on that date told the same story. The median entry lived 144.5 hours, close to six days. 22.6 percent died inside the first hour. 39.0 percent lasted between one day and seven days. 17.0 percent were still up after a month.
Being alive is not the same as being useful. On 5 September 2026 the median live entry passed only 45.5 percent of its own checks. Just 401 of 9,452 live entries passed ten checks at 90 percent success. Median latency was 1,050 ms. 24.1 percent answered under 500 ms and 28.1 percent took over three seconds.
A second list is mostly the first list
This is the part that matters if you came looking for another source. On 5 September 2026 our pool drew from 116 separate sources. The average live entry appeared on 17.86 of them. 94.7 percent appeared on more than one.
The same machines get scanned, found and republished by everyone. Adding ProxyScrape to a script that already reads free-proxy-list.net does not double your supply. It widens the pool a little and refreshes it a lot. Freshness is the real gain, not volume.
Building a rotation that survives a thirty percent list
Never hand a raw line from any file to your real job. Check it first, with the same protocol and a real page. Our audit required the reply to carry an exit address that was not ours. An open port proves nothing at all.
Keep a working set, not a list. Pull the file often and re-check what you hold. Every one of our 72,230 rotating entries had been checked within 24 hours on 5 September 2026. 14,762 had been checked within the previous five minutes. That cadence is the whole trick.
Budget your retries against the target, not against the proxy. If a page fails twice on two different exits, the page is the problem. Cap the connect timeout near three seconds and dial several at once. Drop an entry after two failures and never look back.
Then set your hopes correctly for what a free exit can do. On 5 September 2026, 89.3 percent of our live entries with fraud data were flagged. The average fraud score was 72.5, and 82.9 percent sat on data center address space. None were residential. Free proxies read public pages well. They fail logins, checkouts and anything that scores the visitor.
Method and limits
We fetched the machine-readable file that each list publishes, on 9 September 2026. We parsed every line it held. Then we took a random sample of up to 120 entries. We opened each one, spoke its protocol and fetched a real page through it. The reply had to carry an exit address that was not ours. We ran the whole test twice, thirty minutes apart.
The limits are real. This is one vantage point, one data center in Germany, one day. Your own network will see different numbers. A sample of 120 near a 17 percent rate carries about 3.4 points of error. We measured what each file returned to us, not what anyone intended. ProxyScrape did well in this test and we have no reason to doubt it.
Sources
- HProxy free proxy market audit, 9 September 2026. Eleven lists parsed and sampled. Up to 120 random entries each, two runs thirty minutes apart, from one data center in Germany.
- HProxy free proxy pool, queried 5 September 2026. 763,409 entries seen since launch, 115,078,758 checks recorded, 9,451 alive at that query.
- Lists named here: ProxyScrape, free-proxy-list.net, ProxyNova, Spys.me, Geonode and FreeProxy.World. The GitHub lists: monosans, iplocate, TheSpeedX, Proxifly and clarketm.
If your script needs an exit that answers every time, our paid pools start at $0.44/GB.