Web Data Scraping Benchmarks

Methodology

The index re-uses the raw results of every web data benchmark we published. A benchmark counts if it tests web data providers and records the outcome of each request. A provider's index score is the average of its success rates across the benchmarks it ran in. Every benchmark carries equal weight, so a 1,400-request study counts as much as a 260,000-request one. Success means the response returned the target page with the content we asked for. We check that against bot pages and empty shells, so a 200 response with a challenge page counts as a failure. If a provider was never tested in a benchmark, we show a dash instead of a zero.

Targets come from the Tranco list, which ranks domains by averaging several traffic rankings. We remove dead, adult, gambling and malicious hosts before testing. Domains are grouped into categories: e-commerce, social media, search engines, video and streaming, travel, reviews, food delivery, app stores, real estate, job postings, news and publishing, developer and SaaS, finance, and government and education. Domain coverage and structured JSON support are checked on the top 100. Cost is per 1,000 successful pages, not per 1,000 requests sent. We take the cheapest plan a provider offers at that volume and divide by the success rate we measured. Completion time is how long one request takes from start to finish. Metadata fields are the parsed fields a provider returns when it answers with JSON. AI ready is the share of responses that came back as clean markdown. The anti-bot results come from two of our runs, the unblocker benchmark across 10,000 domains and the e-commerce benchmark across the top 100 domains, with each domain labelled by the vendor guarding it. Whiskers are 95% confidence intervals.