// benchmark

How often, how fast,
how much.

Success rate, latency and cost per 1,000 pages for Scrape and SmartScraper, on 21 public pages across 5 sites you can run yourself.

~/wsai — benchmark2026-10-23
Scrape
▸success100.0%
▸p50 / p950.2s / 0.9s
▸cost / 1k pages1,000 credits
SmartScraper
▸success100.0%
▸p50 / p951.2s / 4.2s
▸cost / 1k pages5,000 credits
210 runs per product · p50 / p95 over successful runs
// method

5 sites, 21 pages. 10 runs each.

The same fixed set every time, so results compare run to run: 210 requests per product, sent one at a time and back to back, from a server in Finland, at each site's own crawl-delay. Caching was off for the run, so every request fetched and extracted its page fresh. Each site is a scraper template; run any of them in the playground.

Books to Scrapehttps://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html5 pages run it
Hacker Newshttps://news.ycombinator.com/1 page run it
Wikipediahttps://en.wikipedia.org/wiki/Web_scraping5 pages run it
GitHubhttps://github.com/nuxt/nuxt5 pages run it
Yahoo Financehttps://finance.yahoo.com/quote/AAPL5 pages run it
All 21 pages
  • https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html
  • https://books.toscrape.com/catalogue/tipping-the-velvet_999/index.html
  • https://books.toscrape.com/catalogue/soumission_998/index.html
  • https://books.toscrape.com/catalogue/sharp-objects_997/index.html
  • https://books.toscrape.com/catalogue/sapiens-a-brief-history-of-humankind_996/index.html
  • https://news.ycombinator.com/
  • https://en.wikipedia.org/wiki/Web_scraping
  • https://en.wikipedia.org/wiki/Web_crawler
  • https://en.wikipedia.org/wiki/Data_scraping
  • https://en.wikipedia.org/wiki/Search_engine
  • https://en.wikipedia.org/wiki/API
  • https://github.com/nuxt/nuxt
  • https://github.com/vuejs/core
  • https://github.com/golang/go
  • https://github.com/microsoft/vscode
  • https://github.com/facebook/react
  • https://finance.yahoo.com/quote/AAPL
  • https://finance.yahoo.com/quote/MSFT
  • https://finance.yahoo.com/quote/NVDA
  • https://finance.yahoo.com/quote/GOOGL
  • https://finance.yahoo.com/quote/AMZN
// per site

Every site, both products.

A slow or difficult site shows up here instead of hiding inside the overall p95.

SiteScrape success · p50 / p95SmartScraper success · p50 / p95
Books to ScrapeScrape100.0%0.2s / 0.2sSmartScraper100.0%1.1s / 2.4s
Hacker NewsScrape100.0%0.3s / 0.8sSmartScraper100.0%4.9s / 8.1s
WikipediaScrape100.0%0.2s / 0.6sSmartScraper100.0%1.3s / 2.1s
GitHubScrape100.0%0.2s / 0.8sSmartScraper100.0%0.7s / 1.1s
Yahoo FinanceScrape100.0%0.8s / 1.0sSmartScraper100.0%1.3s / 1.5s
// what we measure

Three numbers, defined.

Success rate

The share of runs that came back 2xx with the page's data in them. A SmartScraper result must pass the schema sent with the request; a Scrape result must contain the page's own text, so a block page counts as a failure. Requests that return an error cost nothing.

Latency

Wall-clock time at the client, from sending the request to reading the whole response, over successful runs. Requests ran one at a time, back to back. Reported as p50 and p95, not averages.

Cost

Credits for 1,000 successful pages at the base rate. Straight from the price list, so this column is exact today.

~/wsai — try

Run the same pages yourself.

The playground is free and needs no signup: 5 runs every 15 minutes.