ScraperAPI vs ZenRows vs ScrapingBee: which to use in 2026

The agreement floor: 19 of 100

In the October 30, 2026 benchmark run, of the 100 sites, 19 came back with all five attempts from all three providers, among them linkedin.com, zillow.com, walmart.com, amazon.com and nytimes.com. None of the 19 sits inside ZenRows' failure.

On 30 of the 100 sites, exactly one of the three returned all five attempts and neither of the other two managed a full set. ScrapingBee is the sole provider on 21 of those, ZenRows on 5, ScraperAPI on 4. Fourteen of the 30 sit inside ZenRows' failure, so ZenRows never had a working attempt at them. Only temu.com shut out all three. On 14 sites none of the three returned a complete set: temu.com, mouser.com, lowes.com, yelp.com, us.louisvuitton.com, booking.com, monster.com, realtor.com, coupang.com, stockx.com, cargurus.com, tiktok.com, skyscanner.net and allmenus.com.

ScraperAPI and ScrapingBee both cleared 47 sites in common, ScraperAPI and ZenRows 20, and ZenRows and ScrapingBee 27. Running ScrapingBee and ZenRows together reaches 82 of 100 sites at five of five, 6 more than ScrapingBee reaches alone, and ScrapingBee with ScraperAPI reaches 81. All three together reach 86.

In August this section found almost no agreement and argued that no ordering of the three survives a real target list. In September ScraperAPI led across most of the suite. In October ScrapingBee does, and the disagreement still sits in a band of about 30 sites. That band is wide enough that a 20-site list can come out in any order, and the leader changed between two runs a month and a half apart.

ScraperAPI: second of the three, at its 30-credit setting

Benchmark success: 76.4%, 382 of 500 requests, ninth of seventeen providers on the October 30, 2026 run, down from 84.0% and third place in September. 52 sites at five of five, 39 partial, 9 shutouts.

Pricing: Hobby is $49 a month for 100,000 credits and Startup is $149 for 1,000,000, from the pricing page with the toggle on monthly. Startup is the plan nearest $100. Annual billing takes 10% off and the page shows both figures, which is more honest than the alternative discussed below. A standard page is 1 credit; Amazon, Walmart and eBay are 5; Google and Bing and their subdomains are 25; LinkedIn is 30, per the credits documentation. The ultra_premium setting this run used costs 30 credits a request, or 75 with rendering.

Latency: on the 48 sites where String and ScraperAPI both returned all five attempts, ScraperAPI's median was 3.75 seconds against String's 1.70. Its failure-aware latency score, 15.19 seconds, was sixth of seventeen and the best of the three.

Strengths: its 4 sole wins are neimanmarcus.com, aa.com, google.com and redfin.com, two of them inside ZenRows' failure. It leads the three on Akamai at 81.0% and AWS WAF at 76.7%, and it returned all five attempts on marketwatch.com, barrons.com and yellowpages.com where String returned four, and on kroger.com where String returned two. It also documents its costs better than most: there is a URL cost endpoint you can call before you scrape, and a cost control page covering spend caps.

Weaknesses: credits do not roll over. Its own pricing FAQ says the balance resets at renewal, so unused volume is money gone, and the credit usage view is where you find out. It returned nothing on temu.com, mouser.com, realtor.com, stockx.com, capterra.com, instagram.com, facebook.com, x.com and goodrx.com, and it left 39 sites partial, the most of the three. The setting that produced 76.4% is the expensive one, at 30 credits a request. And there is a contradiction inside its own documentation worth knowing before you model costs on the default path. The pricing FAQ says sites behind Cloudflare, DataDome or PerimeterX "add 10 credits per request", which makes a protected page 11 credits. The docs table heads its column "Cost per Scrape", gives 10, and lists Akamai and Cloudflare Turnstile at the same rate. We quote 10 here on the grounds that the docs page is the more specific surface, and flag that our own ScraperAPI alternatives page also gives the 11 reading.

Billing on failure: ScraperAPI charges only for 200 and 404 responses, plus requests you cancel before giving it 70 seconds. That is success billing, so its rate card is already the cost per usable page and should not be divided by 76.4%.

ZenRows: an API failure sits on its number

Benchmark success: in the October 30, 2026 run, 41.8%, 209 of 500 requests, fifteenth of seventeen, with its API closing the connection on every attempt for the last 44 targets. 33 sites at five of five, 14 partial, 53 shutouts, 44 of which are that failure.

Pricing: read the docs plan table, not the pricing page. The pricing page opens on the annual toggle and shows the entry rung of each tier at its annual-equivalent monthly price: $16, $58, $166, $458. The docs give the monthly-billed figures: $19, $29 and $39 across Build, then $69, $99 and $129 across Launch, then $199 upward through Growth. The gap is the pay-for-ten-get-twelve annual discount its FAQ describes. Launch L2 at $99 a month for 500,000 credits is the plan nearest $100 and the closest match to the anchor of any vendor here.

Its credit weights are unusually clear and unusually rigid: 1 credit standard, 5 for JavaScript rendering, 10 for premium proxies, 25 for both together. The docs state that these "are fixed and never change" on every plan, which means moving up a rung buys volume and nothing else. Residential bandwidth meters at 25,000 credits per gigabyte, and the docs give the conversion outright: one gigabyte is 1,000 protected pages. Concurrency moves by tier rather than by rung, so Launch L2 runs 50 parallel requests whichever rung you sit on.

Latency: on the 31 sites where String and ZenRows both returned all five attempts, ZenRows' median was 12.60 seconds against String's 1.70.

Strengths: on the 56 targets its API answered, it passed more attempts than ScraperAPI, 209 against 202. Its 5 sole wins are allegro.pl, indeed.com, safeway.com, g2.com and reuters.com, and it returned all five attempts on marketwatch.com and barrons.com where String returned four. It beats ScraperAPI on Kasada, 60.0% against 53.3%, and as of the September run it was the only one of the three whose benchmark configuration had not changed since July. Its billing is clean: never charged for failures, with 404 and 410 counted as successes.

Weaknesses: outside the failure it still shut out 9 sites: mouser.com, lowes.com, autozone.com, us.louisvuitton.com, monster.com, bloomberg.com, stockx.com, cargurus.com and instagram.com. Its anti-bot columns read 17.8% on in-house protection, 33.3% on Cloudflare and 57.0% on Akamai, but every column includes the 44 failed targets, so treat them as floors rather than measurements. Its flaky middle is 14 sites returning some attempts but not all, including alibaba.com, ebay.com and booking.com at four of five and yelp.com at two. A site that returns four times in five corrupts a daily table more quietly than a site that fails outright, and it bills you for the four.

One thing that changes the cost, not the ceiling: ZenRows ships an Adaptive Stealth Mode (mode=auto) that picks the protection level for you. It changes what you pay per page. It does not change which pages come back, and our run already sent the top configuration by hand.

ScrapingBee: the cheapest credits, and first of the three in October

Benchmark success: in the October 30, 2026 run, 85.2%, 426 of 500 requests, fifth of seventeen, up from 73.0% and seventh place in September. 76 sites at five of five, 15 partial, 9 shutouts.

Pricing: Hobby $19 for 75,000 credits, Freelance $49 for 250,000, Startup $99 for 1,000,000, Business $249 for 3,000,000, Business+ $599 for 8,000,000, from the pricing page. Startup at $99 is the anchor plan and the cheapest credits on this page at $0.099 per thousand. No annual toggle is published, all prices exclude VAT, and the page states that credits are valid for one month and do not roll over.

Then read the credit table before you believe that number. A classic proxy without JavaScript rendering is 1 credit, but rendering is on by default, so the default request is 5. Premium proxy without rendering is 10, with rendering 25, and a stealth proxy with rendering is 75. Its own FAQ confirms the default: "each request costs 5 credits because JavaScript rendering is enabled by default." A team that never sets render_js=false pays $0.495 per thousand standard pages, five times the headline.

Latency: on the 72 sites where String and ScrapingBee both returned all five attempts, ScrapingBee's median was 5.05 seconds against String's 1.70. Its failure-aware latency score, 20.30 seconds, was eleventh of seventeen.

Strengths: Auto Mode is the best-designed billing feature of the three, and it is what this run sent. Add mode=auto and ScrapingBee tries configurations from cheapest to most expensive, stops at the first that works, and charges only for that one. If every configuration fails, the request costs zero credits, and the charge comes back in an Spb-auto-cost response header so you can audit it. max_cost caps how far it is allowed to climb. On a target list with a wide spread of difficulty this is straightforwardly better than paying a flat protected rate on every request. Its 21 sole wins include bestbuy.com, homedepot.com, ebay.com, instagram.com, facebook.com, airbnb.com, seatgeek.com, footlocker.com, wsj.com and alibaba.com, and 12 of the 21 sit inside ZenRows' failure. It reads 92.5% on social media against ScraperAPI's 60.0%, and it leads the three on DataDome, Cloudflare, PerimeterX, Kasada and in-house protection. It also ships a Markdown output mode, which matters if the pages are going into a model rather than a database.

Weaknesses: 9 shutouts: temu.com, lowes.com, autotrader.com, booking.com, realtor.com, reuters.com, coupang.com, google.com and skyscanner.net. It reads 74.0% on Akamai, the largest protection group in the suite, against ScraperAPI's 81.0%, and 63.3% on AWS WAF. Its latency score trails ScraperAPI's, 20.30 seconds against 15.19, and 41 of its 2xx responses were not the requested page.

Where each one wins, by anti-bot vendor

The overall ordering does not fully survive this table. Figures are success rate on the sites carrying each protection, from the October 30, 2026 run. The protections themselves are Cloudflare, DataDome, Akamai Bot Manager, PerimeterX, now sold by HUMAN Security, and Kasada. ZenRows' column includes the 44 targets its API never answered.

Akamai and AWS WAF are the rows to look at. ScrapingBee leads ScraperAPI by 8.8 points overall, and on those two rows, 26 sites between them, ScraperAPI is ahead: 81.0% against 74.0% on Akamai and 76.7% against 63.3% on AWS WAF. ScrapingBee leads or ties every other row, by its widest margin on Cloudflare, 97.3% against 70.7%. If your targets cluster on one protection vendor, the leaderboard is a weak guide and this table is the one to read.

Speed, without the flattery

None of these three is slow in a way that would decide a purchase, and the published latency column is not the place to check. That column imputes a stand-in duration for every attempt a provider failed, which means a provider that misses more sites carries time it never actually spent. It is a fair scoring choice, because it refuses to reward a fast failure, but it is not a like-for-like speed measurement and we will not use it as one.

Split the set instead. On the 17 sites all four of us returned in full, String's median was 1.70 seconds, ScrapingBee's 3.60, ScraperAPI's 4.20 and ZenRows' 12.60. ScrapingBee is the quickest of the three on agreed pages in this run; in September it was ScraperAPI.

Why the pricing does not compare cleanly

Six reasons, all of them visible in the vendors' own documentation. Prices come from each vendor's pricing page and docs, checked October 27, 2026.

1. Two of the three publish credits, not prices, and the weights differ. A protected page is 25 credits at ZenRows and ScrapingBee, and 10 at ScraperAPI. The credit is not a unit.
2. The default configuration is not the cheapest one. ScrapingBee bills 5 credits by default because rendering is on. Its own headline rate assumes you turned that off.
3. One of the three shows annual-equivalent prices as monthly. ZenRows' pricing page renders $58 where the docs say $69, and $166 where the docs say $199.
4. Rollover differs and nobody leads with it. ScraperAPI resets your balance at renewal, ScrapingBee credits last one month, and ZenRows plan credits expire at every renewal, with only top-up credits carried forward once. String rolls credits over.
5. The expensive tiers are the ones that matter and the cheap ones set the headline. Every published per-credit rate is calculated on a standard page. The pages you are buying an API for are the protected ones.
6. Success billing is now the norm, so the old correction is wrong. All three of these bill only for successful responses. Dividing a rate card by a success rate, which is the right move for an attempt-billed vendor, produces a fictional number for all three of these.

What $100 a month buys

At each vendor's anchor plan, using each vendor's own documented weights. All four are success-billed, so no figure below is adjusted by a success rate.

ScraperAPI is the cheapest protected page of the three on published rates, and String undercuts it at this anchor, $0.90 per 1,000 (as of October 2026) against $1.49. Standard pages go the other way: ScraperAPI's $0.15, and ScrapingBee's $0.10 with rendering off, are below String's $0.16. Where String earns that gap is the shutout count, 0 against ScraperAPI's 9 and ScrapingBee's 9, and where it does not, ScraperAPI is the cheaper buy for easy pages. Say that out loud rather than hiding it, because a reader who runs 20 easy sites should buy on price and will find this out anyway.

One more comparison worth having, because it is the one that matches the success rates above. At the configuration our harness sent, ScraperAPI's ultra_premium costs 30 credits, which is $4.47 per 1,000 on Startup. ZenRows' protected pair costs 25 credits, $4.95 per 1,000 on Launch L2. ScrapingBee's Auto Mode charges the winning tier and nothing at all when everything fails, up to $7.43 per 1,000 at the stealth ceiling. Against all three, String's $0.90 per 1,000 premium fetches is the cheaper line.

Where String fits, and where it does not

String is a web access API for teams whose target pages block standard scrapers, that returns the page from one HTTP call. In the October 30, 2026 benchmark run, it returned all five attempts on 95 of the 100 sites and nothing on none. Its flaky middle is 5 sites: marketwatch.com, barrons.com, skyscanner.net and yellowpages.com at four of five, and kroger.com at two. ScraperAPI and ScrapingBee each returned four of those five on every attempt, all but skyscanner.net, and ZenRows returned marketwatch.com and barrons.com in full. That is the clearest case on this page for testing a vendor against a specific list rather than reading a leaderboard.

On the 30 sites where exactly one of the three returned a full set, String returned a full set on all 30. On the 14 where none of the three did, String returned a full set on 13; skyscanner.net is the exception. That is the argument for a single unblocker over a portfolio, and it is a coverage argument, not a price one. If your list is 30 sites and one vendor already returns all of them, this page has not given you a reason to move.

Pricing is on our pricing page, the first 20,000 standard fetches are free, and billing is per successful request.

How to settle this on your own list

Do not take the counts here. Take the method.

1. Pull your actual target list, not a sample of easy pages. 20 URLs is enough to separate these three, because the disagreement sits on exactly the sites people buy an API for.
2. Get the free tiers. ScraperAPI gives 1,000 credits a month, plus a 7-day trial with 5,000. ScrapingBee gives 1,000, ZenRows gives 5,000 a month with no card. String gives 20,000 standard fetches.
3. Run each URL five times, not once. Every partial result on this page would have read as a pass or a fail on a single attempt, and the partials are where the money goes.
4. Check the body, not the status code. All three return 200 on challenge pages. Look for a string only the real page carries.
5. Turn each vendor's strongest setting on, and price that setting rather than the headline. ultra_premium for ScraperAPI, premium_proxy with js_render for ZenRows, mode=auto with a max_cost you have chosen for ScrapingBee.
6. Count shutouts separately from partials, and separate both from API errors. A site that never returns is a data gap you can plan around. A site that returns four times in five is a data quality problem you will find in a dashboard three weeks later. An API that stops answering mid-run is a third problem, and it looks like the first.

Our harness does all six and takes your keys. Fork it and swap the target list.

FAQ

ScraperAPI vs ZenRows vs ScrapingBee, which should I use?

ScrapingBee, if you have to choose without testing: it returned all five attempts on 76 of 100 sites in our October 30, 2026 benchmark run, against 52 for ScraperAPI and 33 for ZenRows. But the three agreed on only 19 of the 100 sites, and on 30 of them exactly one returned a full set. In September the same question had ScraperAPI as the answer. Which one worked changed by site. Test all three on your own list before committing, because the ranking does not predict your result.

Which of the three is best for Cloudflare-protected sites?

ScrapingBee, clearly: in the October 30, 2026 benchmark run it passed 97.3% across the 15 Cloudflare sites in the suite, against ScraperAPI at 70.7%. ZenRows read 33.3%, a figure that includes the targets its API never answered. For comparison, String read 98.7% on the same 15 sites. If your list is mostly Cloudflare, this is the row that decides it.

Which is cheapest?

ScrapingBee, on published credit rates (each vendor's pricing page, checked October 27, 2026): $99 a month for 1,000,000 credits is $0.099 per thousand standard pages. Two things undercut that. Its default request costs 5 credits rather than 1 because rendering is on unless you disable it, and a protected page costs 25. On protected pages ScraperAPI is the cheapest of the three at roughly $1.49 per thousand on its default path. ZenRows is the most expensive protected page at $4.95 per thousand.

Do any of them charge for blocked requests?

No. ScraperAPI bills only 200 and 404 responses, plus requests you cancel before 70 seconds. ZenRows and ScrapingBee both bill only 200, 404 and 410. This is worth stating because it used to be otherwise, and because it means you should not divide any of their rate cards by a success rate to get an effective cost.

Why did ScrapingBee's score jump?

Because our adapter changed. The August run sent ScrapingBee's default JavaScript path on its standard proxy pool, and it scored 29.7%. Before the September run the harness moved it to Auto Mode with the 75-credit stealth tier as its ceiling, the setting ScrapingBee documents for protected pages, and it scored 73.0%, then 85.2% in the October 30, 2026 benchmark run. Treat the September and October figures as what Auto Mode returned at that ceiling, and the August one as a measure of our configuration.

Is it worth buying two of these three?

Sometimes, on this data from the October 30, 2026 benchmark run. ScrapingBee and ZenRows together reach 82 of 100 sites at five of five, 6 more than ScrapingBee alone, and ScrapingBee and ScraperAPI reach 81. Either pairing has a case if your list includes the sites the second vendor took alone. ZenRows' full contribution to a pair cannot be read from this run, because its API failed on 44 of the targets.

Is String cheaper than these three?

Not on every request. Per each vendor's pricing page, ScrapingBee's and ScraperAPI's standard pages are cheaper than String's $0.16 (as of October 2026) per thousand. On protected pages String is the cheapest of the four, $0.90 against about $1.49 on ScraperAPI's Startup plan, $2.48 on ScrapingBee and $4.95 on ZenRows. The case for String on this page is coverage as much as unit price: 95 of 100 sites at five of five. If your targets are sites all four return, buy on price.

Can I check these numbers myself?

Yes, and you should. The suite, every provider adapter and the raw result file are public at usestring/web-data-frontier-benchmark. The adapters are the part to read first, because they hold the flags each provider was sent, and the commit history shows when each was last changed.

Sources

Pricing checked October 27, 2026. Benchmark figures from the October 30, 2026 run.

Benchmark

• Web Data Frontier Benchmark harness
• Method and latency scoring
• Raw results, October 30, 2026 run
• Adapter change before the September run
• ScraperAPI adapter
• ZenRows adapter
• ScrapingBee adapter
• Benchmark results on usestring.ai
• The October 2026 benchmark post

ScraperAPI

• Pricing
• Credits and request costs
• Premium residential and mobile proxy pools
• Cost control
• Supported parameters
• API status codes
• Credit usage

ZenRows

• Pricing page
• Pricing documentation
• Fetch API reference
• API error codes
• Getting started
• Fetch product page
• Concurrency
• Status page

ScrapingBee

• Pricing
• Documentation, including the credit cost table and Auto Mode
• FAQ
• Knowledge base
• CLI documentation
• Markdown scraper
• Status page

Anti-bot vendors named in this page

• Cloudflare bot management
• DataDome documentation
• Akamai Bot Manager
• HUMAN Security, which now sells PerimeterX
• Kasada

Related String pages

• ScraperAPI alternatives
• ZenRows alternatives
• Zyte alternatives
• Web Access API
• Pricing