String returned verified content on 97.0% of requests in the open Web Data Frontier Benchmark. ScrapingBee returned 29.7%. The August 11, 2026 run covered 99 real sites with five attempts each.
That is a large result, but it needs an equally large caveat. The benchmark used ScrapingBee's default JavaScript-rendered HTML API, not its premium proxy, stealth proxy, or newer Auto Mode. The result measures the default path, not the ceiling of every ScrapingBee configuration. Pick String when protected pages have to come back with no manual proxy tuning. Pick ScrapingBee for lower prepaid pricing, explicit configuration control, dedicated data APIs, and current Auto Mode that can escalate beyond the path tested here.
TL;DR
| String | ScrapingBee | |
|---|---|---|
| Product type | Unblocker: fetch, search, sitemap, remote browser | HTML scraping API plus dedicated search, retail, AI, screenshot, and extraction tools |
| Benchmark success | 97.0% (480/495), 1st of 15 | 29.7% (147/495), 15th of 15 |
| Latency score | 9.98s, 2nd | 16.97s, 10th |
| Sites with no successful attempt | 2 of 99 | 62 of 99 |
| $100-tier billing | Growth: $100 subscription plus $0.20 to $4.00 per 1,000 successful requests | Startup: $99 prepaid, 1,000,000 credits, 100 concurrent requests |
| Failure billing | Blocked requests are free | 200, 404, and 410 responses bill; other failures are free |
| Automatic escalation | Built into the endpoint | Auto Mode chooses rendering and proxy strength, with an optional cost cap |
| MCP and integrations | Hosted MCP registry entry | Hosted MCP, CLI, LangChain, n8n |
| Where it wins | Measured coverage, latency, automatic retrieval on hard targets | Prepaid price, manual controls, Auto Mode, dedicated APIs, SDK and integration breadth |
We built the benchmark and sell the product that ranks first. The test is public so buyers do not have to trust our summary. The 99 targets, methodology, provider adapters, and August results are available to inspect and rerun.
Each provider received the same 99 URLs five times. A response passed only when it contained a marker from the real page. A captcha or challenge page that returned HTTP 200 failed.
The ScrapingBee adapter used the official SDK with render_js: true and json_response: true. It did not set premium_proxy, stealth_proxy, or mode: auto.
That choice reflects ScrapingBee's default HTML API path, because JavaScript rendering is on by default. It is not ScrapingBee's strongest path. ScrapingBee introduced Auto Mode on July 22, 2026, twenty days before the run. Auto Mode can try classic, premium, and stealth proxies with or without rendering, stopping at the first success. This benchmark should add that mode in a future rerun. Until then, read 29.7% as a default-path result, not a claim about what ScrapingBee can do at 75 credits per request.
ScrapingBee is a configurable scraping platform. Its HTML API exposes JavaScript rendering, premium and stealth proxies, country selection, sessions, custom headers and cookies, screenshots, CSS or XPath extraction, AI extraction, and browser actions. It also has a proxy mode for tools that expect an HTTP or SOCKS proxy.
The product surface extends beyond generic page retrieval. ScrapingBee offers Fast Search, Google, Amazon, Walmart, and YouTube APIs, AI and LLM endpoints, a hosted MCP server, and a CLI that can crawl a domain or sitemap. The AI extraction API adds five credits to the underlying scrape and returns structured fields from a natural-language request.
String is narrower. Send a URL to one fetch endpoint and receive markdown, HTML, or JSON. Proxy selection, fingerprints, captcha handling, rendering, and retries happen behind the request. String also has /search, asynchronous sitemap discovery, a remote browser over CDP, and MCP tools. It does not expose ScrapingBee's proxy ladder or its catalog of dedicated retail and search APIs.
String passed 480 of 495 requests. ScrapingBee passed 147, a gap of 333 successful responses across the same target list.
| Per-target outcome | String | ScrapingBee |
|---|---|---|
| Passed all 5 attempts | 92 | 22 |
| Passed 1 to 4 attempts | 5 | 15 |
| Passed no attempts | 2 | 62 |
ScrapingBee's 22 perfect targets include Amazon, Walmart, LinkedIn, Facebook, YouTube, Pinterest, Bloomberg, The New York Times, TradingView, CoinMarketCap, Airbnb, Zara, and Target. It can be a good fit when your workload sits inside that coverage.
The 62 shutouts include Home Depot, Lowe's, Best Buy, AutoZone, AutoTrader, Etsy, eBay, Reddit, GitHub, Stack Overflow, Expedia, Booking.com, Hyatt, American Airlines, Kayak, Realtor.com, Apartments.com, Yelp, G2, Trustpilot, Ticketmaster, StubHub, AXS, and Wall Street Journal. String passed 56 of the 62 on every attempt. Temu and Skyscanner defeated both products, while String passed four of five on Mouser, Neiman Marcus, Allegro, and Booking.com.
ScrapingBee did not beat String on any target in this configuration. That is a benchmark fact, not a claim that Auto Mode cannot win one.
The category split explains the result. ScrapingBee's default JavaScript path was strongest against PerimeterX and self-defended sites, and weakest against Cloudflare and Fastly.
| Anti-bot system | Sites | String | ScrapingBee |
|---|---|---|---|
| Cloudflare | 15 | 100% | 9.3% |
| DataDome | 19 | 97.9% | 29.5% |
| Akamai | 20 | 98.0% | 20.0% |
| PerimeterX | 9 | 88.9% | 46.7% |
| AWS WAF | 6 | 96.7% | 26.7% |
| Kasada | 3 | 100% | 33.3% |
| Fastly | 1 | 100% | 0% |
| No named vendor | 26 | 96.2% | 44.6% |
Premium or stealth proxies are the documented ScrapingBee response to these blocks. The benchmark did not turn them on, so this table is useful for evaluating the default path and for identifying exactly what a rerun with Auto Mode needs to test.
String's latency score was 9.98 seconds, second of 15. ScrapingBee's was 16.97 seconds, tenth. The score uses the per-target p75 of successful attempts and charges a provider for targets it cannot return, so fast failures do not look like speed.
The raw mean attempt time tells the opposite story: 6.59 seconds for ScrapingBee and 7.67 for String. ScrapingBee's many fast failures lower that mean, which is why the benchmark does not rank providers by it. On the 37 sites ScrapingBee reached at least once, its response time may be perfectly competitive. On a fixed 99-site dataset, time to complete is set by the 62 sites that never arrive.
ScrapingBee Startup is $99 per month for 1,000,000 prepaid credits and 100 concurrent requests. Its credit system charges one credit for classic proxies without rendering, five with rendering, ten for premium without rendering, 25 for premium plus rendering, and 75 for stealth.
At full use of Startup, those weights translate to:
| ScrapingBee configuration | Credits/request | Pages included | Price per 1,000 |
|---|---|---|---|
| Classic, no JavaScript | 1 | 1,000,000 | $0.10 |
| Classic with JavaScript | 5 | 200,000 | $0.50 |
| Premium, no JavaScript | 10 | 100,000 | $0.99 |
| Premium with JavaScript | 25 | 40,000 | $2.48 |
| Stealth with JavaScript | 75 | 13,333 | $7.43 |
String Growth is a $100 monthly subscription plus usage, not a prepaid bundle. Usage per 1,000 successful requests is $0.20 for a request path on a standard proxy, $2.00 on a premium proxy, $1.00 for a browser on a standard proxy, and $4.00 for a browser on a premium proxy. Blocked requests are free and credits roll over.
ScrapingBee is genuinely cheaper for several workloads. One hundred thousand classic JavaScript requests fit inside Startup for $99. The same count on String Growth costs at least $120 if every page uses the standard request path, and more if it needs a browser or premium proxy. At 100,000 premium-plus-JavaScript pages, ScrapingBee Business has enough credits at $249, while String Growth plus premium-request usage is $300 for 100,000 successful requests.
String can be cheaper at the strongest edge. One hundred thousand stealth requests need 7.5 million ScrapingBee credits, fitting the $599 Business+ plan. String's most expensive published path, browser plus premium proxy, is $500 total at 100,000 successful requests: $100 subscription plus $400 usage. Whether that path delivers the same page is what the benchmark is meant to answer.
Both products use success-oriented billing, with an important ScrapingBee footnote. ScrapingBee bills HTTP 200, 404, and 410 responses; other failures are free. A dead URL can therefore spend credits. Auto Mode bills only the successful configuration it selects and costs zero credits if all allowed configurations fail. String bills only when verified content comes back.
Pick String when your target list is varied, protected, and incomplete data is the main risk. Pick ScrapingBee when the sites are known to work, prepaid unit cost matters, or you need its dedicated APIs and explicit controls. If you choose ScrapingBee for hard targets, test Auto Mode rather than copying the benchmark adapter. If you choose String, test the exact pages that drive your cost because the usage path can range from $0.20 to $4.00 per 1,000 on top of Growth.
For default-path coverage in this benchmark, yes: String passed 97.0% against ScrapingBee's 29.7%. ScrapingBee has lower prepaid pricing, more manual controls, dedicated data APIs, and Auto Mode that the benchmark did not test.
No. The adapter enabled JavaScript rendering but did not enable premium proxies, stealth proxies, or Auto Mode. The 29.7% result is the documented default path, not ScrapingBee's maximum. A future run should test mode=auto and publish its credit distribution.
Often. ScrapingBee Startup includes 1,000,000 credits for $99, while String Growth is $100 plus usage. At full use, ScrapingBee's classic JavaScript path is $0.50 per 1,000 and premium plus JavaScript is $2.48. String Growth usage ranges from $0.20 to $4.00 per 1,000 successes on top of the subscription.
ScrapingBee bills responses with HTTP 200, 404, or 410 and does not charge other failures. Auto Mode charges only the configuration that works and zero credits if all allowed configurations fail. String charges only when verified page content comes back.
Auto Mode tries classic, premium, and stealth proxy configurations with or without JavaScript rendering, starting with the cheapest. It stops at the first success. max_cost can keep it from trying expensive tiers, and a response header reports the selected credit weight.
Yes. ScrapingBee hosts a remote MCP server for scraping, screenshots, search, and dedicated data APIs. String also hosts an MCP server with fetch, search, and sitemap tools.
Twenty-two targets passed all five attempts, including Amazon, Walmart, LinkedIn, Facebook, YouTube, Pinterest, Bloomberg, The New York Times, TradingView, CoinMarketCap, Airbnb, Zara, and Target. String also passed all of those targets on every attempt.
Run the public benchmark repository with your own API keys. For a current ScrapingBee test, change its adapter to Auto Mode, record the Spb-auto-cost header, and compare both returned content and credits used.
