Last month we published the September edition with 80 providers and 115,192 requests. The October edition is out: 82 providers, 6 targets, 142,310 requests, collected from Oct 1 to Oct 31, 2026 (all dates in this post are UTC).

Two changes affect how you read it. We changed the order in which providers are tested partway through the month, and Google and Amazon results are much higher than in September. We don't know why, and we say so below. At the end there is a short pandas script so you can check the headline numbers against the public dataset.

At a glance

Metric October 2026 vs September
Providers tested 82 +7 new, −5 exited
Test scenarios 6 same
Total requests 142,310 +23.5%
Avg success rate 91.5% +4.0 pp
Avg latency 1.48 s +0.12 s
Overall block rate 6.1% −3.9 pp

Percentage-point changes are computed from the unrounded rates, then shown to one decimal. 91.5 − 87.4 is 4.1, and the published change is +4.0 pp.

1. The target matters more than the provider

Same providers, same month, very different results depending on where the request goes.

Target Requests Success Avg latency P95 Block rate vs Sep
ipify (generic HTTPS) 23,650 98.4% 0.67 s 2.16 s 0.0% +0.3 pp
Nike 23,727 97.7% 2.05 s 4.28 s 0.0% −0.2 pp
Instagram 23,728 97.0% 1.40 s 3.54 s 0.6% +0.3 pp
TikTok 23,727 96.5% 1.38 s 3.22 s 0.0% +0.6 pp
Google 23,745 88.3% 0.75 s 1.21 s 9.2% +11.8 pp
Amazon 23,733 70.9% 3.00 s 5.39 s 26.9% +10.8 pp

The spread from easiest to hardest is 27.5 points. A provider's "average success rate" mostly tells you which targets were in the mix.

What this means for you: test against your own target. A generic connectivity check will make almost any provider look fine.

2. Google and Amazon jumped, and we can't explain it

Google is up 11.8 points and Amazon 10.8 points against the published September numbers. We checked the obvious suspects:

  • Classification rules, targets, and test nodes did not change.
  • The increase was already there in the first week of October.
  • The other four targets moved by less than a point.

We did change the provider test order on Oct 8 (see the methodology notes), but the Google and Amazon levels were already higher before that. We have not identified the cause, so we are not attributing it to anything.

Amazon is also volatile day to day. Overall daily success fell below 90% on Oct 2, 5, 7–9, 13–16, and 26. Those dips line up with days when Amazon success dropped across nearly all providers, while Nike stayed around 96–99%. Amazon's monthly 70.9% is an average of good days and days when almost everyone was blocked.

What this means for you: don't plan capacity from a monthly average on a hard target. Build retries and backoff for whole-day bad stretches, not just single failed requests.

3. The top 10 is tightly packed

Only providers with at least 1,500 requests in the window are eligible: 71 of the 82. Ranking is by weighted success rate, with average latency as the tie-breaker.

# Provider Requests Success Avg latency P95 Block rate
1 KocerRoxy 1,860 96.3% 1.21 s 2.68 s 3.5%
2 EnigmaProxy 1,860 95.7% 1.58 s 3.35 s 3.8%
3 FlashProxy 1,860 95.5% 1.91 s 4.81 s 4.4%
4 ResiProx 1,860 95.4% 1.46 s 2.90 s 4.5%
5 ScrapingAnt 1,860 95.4% 1.47 s 3.63 s 4.4%
6 anyIP 1,860 95.4% 1.49 s 2.87 s 3.8%
7 ProxyEmpire 1,860 95.2% 0.80 s 1.58 s 4.8%
8 ProxyWing 1,860 95.2% 1.26 s 2.76 s 4.3%
9 9HTTP 1,861 95.2% 1.83 s 4.11 s 3.9%
10 Databay 1,860 94.95% 1.22 s 2.46 s 4.5%

The gap from first to tenth is about 1.3 points. At that spread, block rate, latency, and your own target matter as much as the headline number. Databay is shown to two decimals because one decimal would round it up to 95.0%.

Three more providers cleared 95% with fewer than 1,500 requests. We kept them out of the top 10 on purpose, so a short run of lucky requests can't headline the report.

What this means for you: read request counts next to every success rate. A 97% on 300 requests is a weaker claim than 95% on 1,860.

4. High success does not mean low latency

ProxyEmpire was the fastest in the top 10: 0.80 s average and 1.58 s P95 at 95.2% success. FlashProxy held 95.5% success with a 4.81 s P95, three times ProxyEmpire's tail.

What this means for you: if you run with timeouts or synchronous workers, look at P95, not the average.

5. Outages stay in the numbers

We log an endpoint incident when a provider fails every live scenario in a run, even if a later run the same day succeeded. Those failed runs stay in the rankings, because they reflect real availability.

Provider Date Duration Requests affected Success impact
LumiProxy Oct 21 1 day 60 −3.0 pp (95.5% → 92.5%)
YumiProxy Oct 22 1 day 60 −2.9 pp (91.3% → 88.4%)
Coronium Oct 26 1 day 60 −2.8 pp (86.5% → 83.8%)
NSTProxy Oct 28–31 4 days 240 −11.8 pp (91.4% → 79.6%)
Swiftproxy Oct 30–31 2 days 120 −5.4 pp (83.0% → 77.7%)

The NSTProxy and Swiftproxy incidents were still running on the last day of the data window.

6. New and exited providers

New this month: B2Proxy, Geekproxy, Luxproxy, Proxmint, ProxyCove, Thordata, Webshare.

Exited testing: 711Proxy, Anonymous Proxies, Astro, Omegaproxy, ProxySale.

Exited providers keep their earlier data in the cohort totals. We remove data from the exit date onward and we do not rank them.

The cohort changed, and the test order changed. Treat provider-level month-over-month comparisons with caution.

Methodology notes

  • Schedule: one run per day at 18:00 UTC, 10 requests per provider per scenario, through each provider's residential gateway. ISP-only providers are excluded from this edition.
  • Test order: provider order within each scenario was fixed (sorted by slug) through the Oct 7 run and shuffled from Oct 8. September used the fixed order.
  • Weights: success rate and block rate are weighted by all requests. Average latency and P95 are weighted by successful requests.
  • Eligibility: top 10 and month-leader claims need at least 1,500 requests. 71 providers qualify.
  • Independence: affiliate links and partnership status do not change scores.

Full detail is on the How we test page.

Check our numbers

Download the JSON from the bottom of the report page. The file name is proxyvero-2026-10-dataset.json. This script recomputes the per-target success rates from provider_scenario, and the top 10 from the providers list using the same top10_eligible flag as the report.

import json import pandas as pd with open("proxyvero-2026-10-dataset.json", encoding="utf-8") as f: report = json.load(f) cells = pd.DataFrame(report["provider_scenario"]) providers = pd.DataFrame(report["providers"]) def weighted_mean(group, column, weight="requests"): usable = group.dropna(subset=[column]) total = usable[weight].sum() if not total: return float("nan") return (usable[column] * usable[weight]).sum() / total by_target = ( cells.groupby("scenario") .apply(lambda g: pd.Series({ "requests": int(g["requests"].sum()), "success_pct": weighted_mean(g, "success_rate"), })) .sort_values("success_pct", ascending=False) ) print(by_target.round(1)) eligible = providers[providers["top10_eligible"] == True].copy() top10 = eligible.sort_values( ["success_rate", "latency_avg_ms"], ascending=[False, True], ).head(10) top10["latency_s"] = top10["latency_avg_ms"] / 1000.0 print(len(eligible), "eligible providers") print(top10[["provider", "requests", "success_rate", "latency_s"]].round(2).to_string(index=False))

success_rate is already a percent from 0 to 100. latency_avg_ms is milliseconds, and on each provider row it is already weighted by successful requests. The per-target success rates above weight every request, which is the same method as the report. If a number does not match, tell us.

Next edition and your input

The November edition will be published in early December. Which targets or metrics would make this more useful to you: other marketplaces, search engines, per-country results, cost per successful request? Tell us in the comments. The target list for next month is still open.