Web Scraping API Benchmark 2026: Success Rates on Protected Sites
When you evaluate a web scraping API or web unblocker, it is easy to get lost in feature lists, pricing tables, and marketing claims. But there is really only one number that decides whether a provider is worth your money: how often does it actually return the real page from a protected site? Everything else is secondary. A cheap API that fails half the time on the targets you care about is not cheap, it is a slow, expensive way to collect nothing.
This article does two things. First, it walks through what independent benchmarking looks like, using Proxyway’s 2025 Web Scraping API Report as a reference point, because their methodology and published numbers are some of the most rigorous public data available. Second, we took Proxyway’s exact target list and ran it ourselves against our production API, a full benchmark of 6,000 calls per target at concurrency 10, scored by page content, and put our numbers side by side with theirs. You can read the complete write-up, all 14 targets with 6,000 calls each, on our full benchmark page.
Why Success Rate on Protected Sites Is the Metric That Matters
Most websites are not hard to scrape. A blog, a small e-commerce shop, a public dataset, most of these will hand you their HTML with nothing more than a plain HTTP request. You do not need a benchmark to tell you that.
The difficulty, and the reason web unblocker APIs exist at all, lives in a much smaller set of high-value targets sitting behind serious anti-bot infrastructure. Think large retailers, travel and hospitality sites, review platforms, and marketplaces. These sites deploy commercial bot-detection systems that fingerprint your TLS handshake, inspect your browser’s JavaScript environment, score your IP reputation, and throw JavaScript challenges or CAPTCHAs at anything that looks automated. The vendors you run into most often are Cloudflare, DataDome, PerimeterX, Akamai, Incapsula, Kasada, and Shape.
Against these systems, raw proxies alone are not enough. You need a managed pipeline that combines residential or mobile IPs, a real browser environment, correct fingerprints, header and cookie handling, and automatic retry logic, all tuned per target. That is what a web unblocker API is supposed to do for you. So the honest way to compare providers is not to count features, it is to point them at genuinely protected sites and measure the success rate.
What “Success” Actually Means
This sounds obvious but it trips up a lot of naive tests. An HTTP 200 response is not success. Anti-bot systems love to return a 200 status code with a challenge page, a CAPTCHA, or an “Access Denied” body. If you only check the status code, you will happily record a block as a win and your benchmark becomes fiction.
Real success means the response body contains the actual page content you asked for: the product listing, the review text, the search results. In our own benchmark below, a call counted as a success only when the response body carried the genuine page and did not match a challenge or “block-as-200” page. That makes our numbers conservative, not raw HTTP 200 rates. It is the same spirit as a serious benchmark: you check what came back, not just the status line.
How Proxyway Runs Its Benchmark
Proxyway is an independent research site that publishes recurring benchmarks of proxy and scraping-API providers, and their reports are widely cited across the industry. Their 2025 Web Scraping API Report is a good example of what defensible testing looks like, so it is worth understanding their approach before looking at any numbers.
According to Proxyway’s published methodology, the 2025 unblocking test covered 11 scraping APIs against 15 heavily protected websites. They collected roughly 6,000 unique URLs per target and fetched them at 2 requests per second and again at 10 requests per second, with a 600-second timeout, running the bulk of the tests in October 2025. Testing at two request rates is important because plenty of providers look strong at a gentle pace and then fall apart under load, and a real benchmark surfaces that difference.
They also documented which targets were hardest. Averaged across all providers at 2 req/s, Shein returned successfully only 21.88% of the time, G2 36.63%, and Hyatt 43.75%, according to Proxyway’s report. In other words, even the best commercial tooling struggles badly with a handful of the most aggressively defended sites, and no honest vendor should claim otherwise.
How We Ran the Same Targets
Since Proxyway does not list ScrapeUnblocker, we took their target list and ran it ourselves against our production API (api.scrapeunblocker.com). Every target was measured with 6,000 calls at concurrency 10, and every response was content-verified: a challenge page or a “block-as-200” body counts as a failure, not a success. We ran 14 of Proxyway’s 15 targets.
Two things to keep in mind before the numbers. Proxyway ran their tests at 2 requests per second and we ran at concurrency 10, and the two are not a shared execution, so treat every side-by-side here as directional, not a decimal-accurate ranking. And the Proxyway “field average” columns below are the average across Proxyway’s 11 tested providers, which makes each one a difficulty measure for that target, not any single competitor.
Per-Target Results
Here is each of Proxyway’s targets, its protection vendor, our production success rate over the full 6,000-call run, our average response time, and the Proxyway field average. Delta is our rate minus that field average.
| Target | Protection | Our success | Avg response | Proxyway field avg | Δ vs field |
|---|---|---|---|---|---|
| shein.com | in-house | 99.3% | 7.4s | 21.88% | +77.4 |
| g2.com | DataDome | 99.9% | 12.1s | 36.63% | +63.3 |
| hyatt.com | Kasada | 99.9% | 19.8s | 43.75% | +56.2 |
| immobilienscout24.de | Incapsula | 99.7% | 12.2s | 71.68% | +28.0 |
| walmart.com | PerimeterX+ | 100.0% | 6.5s | 93.05% | +7.0 |
| google.com | in-house (serpApi) | 94.4% | 9.3s | 94.78% | -0.4 |
| leboncoin.fr | DataDome | 99.6% | 12.3s | 63.83% | +35.8 |
| allegro.pl | DataDome | 95.5% | 14.0s | 66.98% | +28.5 |
| youtube.com | in-house | 98.0% | 10.1s | 93.05% | +5.0 |
| zillow.com | PerimeterX | 99.6% | 17.1s | 97.85% | +1.8 |
| instagram.com | in-house | 86.7% | 6.4s | 59.54% | +27.2 |
| amazon.com | in-house | 100.0% | 3.5s | 93.30% | +6.7 |
| lowes.com | Akamai | 99.4% | 28.3s | 52.57% | +46.8 |
| nordstrom.com | Shape | 86.5% | 15.5s | 61.97% | +24.5 |
The two gaps that matter most sit at the top: on Shein, where the field averages just 21.88%, we ran 99.3%, and on G2 (DataDome), a 36.63% field average, we ran 99.9%. Those +77 and +63 point gaps are the real story. Lowe’s (Akamai) is a similar picture, 99.4% against a 52.57% field. Every one of the 14 targets we ran carries a Proxyway field average, so a delta is shown for all of them.
Overall Success vs Each Provider
Our overall average sits next to the overall success Proxyway reports for the six general-purpose providers across their 15 hard targets. Ours is 14 of 14 targets finished, 6,000 calls each at concurrency 10, content-verified, each target weighted equally. Treat this table as directional, not an apples-to-apples run.
| Provider | Overall success | Avg response |
|---|---|---|
| ScrapeUnblocker (this benchmark, 14 of 14 targets) | 97.0% | 12.5s |
| Zyte | 93.14% | 11.15s |
| Decodo | 85.88% | 15.22s |
| Oxylabs | 85.82% | 16.76s |
| ScrapingBee | 84.47% | 24.46s |
| ZenRows | 70.39% | 19.10s |
| ScraperAPI | 68.95% | 13.92s |
Our figure is the mean of per-target averages over the finished 6,000-call runs. Proxyway measures its own average response differently, so read the response column as an order of magnitude rather than a decimal-accurate ranking.
Web Unblocker Field and Price
Proxyway also runs a separate, lighter “web unblocker” test set. Here it is alongside published pricing, with our row for reference. The success numbers come from that separate lighter test, so they are not directly comparable to the tables above, the useful column here is price.
| Product | Success (light set) | Price |
|---|---|---|
| ScrapeUnblocker | 97.0% | from 0.55€ / 1,000 (~$0.60/1k) |
| Oxylabs Web Unblocker | 98.5% | $15 / GB |
| Bright Data Web Unlocker | 97.9% | $3 / 1k |
| Decodo Site Unblocker | 96.3% | $34 / 15k |
| Nimbleway | 95.5% | $3 / 1k |
| SOAX | 68.6% | — |
| Rayobyte | 37.65% | $12 / GB |
Our 97.0% is the overall from the table above, shown for reference; the competitor figures here are from Proxyway’s separate lighter web-unblocker test, so success is not directly comparable. On price, at roughly $0.60 per 1,000 calls (from 0.55€ / 1,000) we are about 5x cheaper than the $3/1k tier such as Bright Data and Nimbleway, and far below usage-metered $/GB pricing like Oxylabs at $15/GB.
Reading the Results Honestly
A few important caveats, stated plainly, because a benchmark you cannot trust is worse than no benchmark.
- Field averages are a difficulty measure, not a single competitor. A target where we sit below the field (see the delta column) is one where Proxyway’s providers, at 2 req/s, did better than we did at concurrency 10. Google is the clearest example, a wash at -0.4.
- Strict scoring. Success is content-verified, so challenge pages and “block-as-200” bodies count as failures. These are conservative numbers, not raw HTTP 200 rates. Proxyway ran at 2 req/s; we ran at concurrency 10.
- Not identical test sets. The provider comparisons in the tables above are against Proxyway’s own runs, not a shared execution, so treat them as directional.
- We ran 14 of Proxyway’s 15 targets. The one target we exclude has an average response above the 60-second timeout used, so a like-for-like run scores it 0%. Excluding it flatters us slightly, since Proxyway’s providers were scored on it. We do not add targets of our own.
- We fixed our own mistakes before publishing. An earlier version of the run was dropped because several figures did not survive scrutiny: the harvest fed junk URLs to some targets (analytics hosts picked up from preconnect hints, a dead seed URL, an image CDN), a concurrency bug in our render path briefly let parallel calls return each other’s pages, and an Amazon Akamai interstitial was being counted as a success. All three were fixed, and every row above is from the corrected 6,000-call pipeline.
- We cite Proxyway as a source, we do not reproduce or impersonate it. Every Proxyway figure here comes from one report, the Web Scraping API Report 2025. Proxyway also publishes a separate 10-target run at 10 req/s where the same sites score differently (G2 at 60.39% rather than 36.63%); nothing here is taken from it.
Where ScrapeUnblocker Fits
ScrapeUnblocker is a managed web unblocker API: you send a URL, and it handles the residential and mobile IPs, the real browser rendering, the fingerprinting, the headers and cookies, the anti-bot challenges (Cloudflare, DataDome, PerimeterX, Akamai, Incapsula, Kasada, Shape), and the retries, then returns the finished HTML or AI-parsed JSON. The goal is the same one every number in this article is really about, getting the real page back from sites that are built to keep bots out.
The story these numbers tell is not that we win everywhere. Google is a wash, and instagram and nordstrom are our two softest targets in the mid-80s. It is that on the industry-hardest anti-bot targets, the ones where the field collapses into the 20s, 30s, and 40s, we hold up: Shein at 99.3% against a 21.88% field, G2 at 99.9% against 36.63%, Lowe’s at 99.4% against 52.57%. Overall we measured 97.0% across all 14 targets, at roughly a fifth of the price of the $3/1k tier.
The Takeaway
If you are choosing a web scraping API, do not buy on feature lists or a single headline percentage. Do this instead:
- Test on your own targets. The only success rate that matters is the one measured on the specific protected sites you need, at the request rate you will actually run.
- Define success as real content, not HTTP 200. Verify the body.
- Test under load, not just a handful of polite requests, because that is where providers separate.
- Trust dated, methodology-first benchmarks like Proxyway’s for the broad landscape, and treat any vendor’s self-reported number, including ours, as a starting point to verify, not gospel.
We would rather show you every target, drop the numbers from a run that did not survive scrutiny, and tell you exactly where we sit below the field than publish a rounded-up number you could never reproduce. Success rate on protected sites is the metric that matters, and the only benchmark you should fully trust is the one you run yourself. Point ScrapeUnblocker at your hardest targets and check the content that comes back, or read the full benchmark first, all 14 targets with 6,000 calls each.
Proxyway figures are from Proxyway’s 2025 Web Scraping API Report; the field averages are the mean across Proxyway’s 11 tested providers and are a difficulty measure, not a single competitor. ScrapeUnblocker figures were measured on our production API in July 2026; see the full benchmark for all 14 targets. Proxyway did not test ScrapeUnblocker, and this article cites Proxyway as a source and is not affiliated with or endorsed by Proxyway. Trademarks belong to their respective owners. This is a directional comparison, not a controlled head-to-head, and pricing is quoted as published at the time of writing.
Try ScrapeUnblocker free
95%+ success rate · from 0.55€ per 1,000 calls · 500 free requests on signup.