Which web scraping tool costs the least per successful page in 2026?

Apify runs "Actors": prebuilt scrapers published in a public Store that you launch by filling in a form. For a popular site there is almost always an Actor already. Paste a URL, set how many results you want, and download CSV or JSON.

The bill has two layers. Your plan sets a monthly credit and a compute unit (CU) price: $0.30 on Free, $0.25 on Scale, $0.20 on Business. On top, many Actors charge per result at a price their developer chooses. For Google Maps, that ranges from $1 to $9.99 per 1,000 results depending on which Actor you pick. Two Actors that do the same job can differ tenfold.

Users like the per-run spending caps, which stop a runaway job from draining your balance. The complaints are consistent too: community Actors vary a lot in quality, finding the right one takes time, and pay-as-you-go "adds up" once proxies and storage, billed separately, join the party.

Tip: before you order 10,000 results, run the Actor on 100 and read the cost in the run history. Multiply by 100. Then try a second Actor for the same site and compare.

Firecrawl turns a web page into clean markdown (the format language models read best) or into structured JSON that follows a schema you define. It can also crawl a whole site from one starting URL. Reach for it when the data ends up in an LLM rather than a spreadsheet.

Its grid is one of the easiest to read: 1 credit per page. Hobby gives 5,000 credits for $19, Standard 100,000 for $99, Growth 500,000 for $399. Two catches. JSON output costs 4 extra credits per page, 5 in total. And unused credits expire at month end, except on Scale (one month) and Enterprise.

There is an open source version you can self-host. On GitHub, several issues describe the self-hosted build getting blocked by Cloudflare on sites the cloud version handles. The managed anti-blocking layer is what you pay for, and the free build does not include it.

Tip: if you only need the text, stick to markdown. You pay a fifth of the JSON price, and an LLM can pull fields out of markdown perfectly well afterwards.

ScrapingBee is an API. Your code sends a URL; you get back the page HTML, rendered in a real browser when needed, fetched through a proxy it picks. You keep control of parsing and offload the part that breaks most often.

Mind the multiplier. A plain request costs 1 credit. JavaScript rendering, 5 credits. Premium proxies, 10 or 25 credits depending on options. So the 250,000 credits in Freelance are 250,000 easy pages, or 10,000 hard ones. On the plus side, the pricing page says you pay for successful requests.

Concurrency runs from 25 parallel requests on Hobby to 100 on Startup. For a small team, Hobby at $19 easily covers 10,000 plain pages a month (10,000 of 75,000 credits).

Tip: don't turn on JavaScript rendering by default. Fetch the page without it first, check whether the data is already in the HTML, and only pay for the browser on pages that really need it. On an e-commerce catalog that's often the gap between Hobby and Startup.

Octoparse is a desktop app. You open the page in its built-in browser, click a price, a title and the "next" button, and it works out the pattern. It also ships templates for common sites. It's the most visual approach in this ranking.

The free plan is real: 10 tasks, one device, runs on your machine, up to 10,000 rows per export and 50,000 rows a month. Cloud runs (on Octoparse servers, laptop closed) start with Standard, from $69 a month billed annually for 100 tasks and 3 parallel cloud processes. Professional is $249 a month billed annually.

The catch is commercial as much as technical. Octoparse's own billing docs say a trial turns paid unless you cancel before it ends, and you must cancel on the website, at least a day before renewal, not through PayPal. On review sites, users describe being charged after a trial and refused a refund. Pagination setup is called "not very intuitive", and some runs come back incomplete.

Tip: stay on the free plan as long as your computer can stay on during the run. If you start the trial, set a reminder two days before it ends.

Browse AI works with "robots" you record by showing it what you want on a page. It then reruns them on a schedule and sends the results to Google Sheets or a webhook. Its sweet spot is monitoring: it tells you when a price or a block of text changes.

The free plan gives 50 credits a month across 2 sites, enough to confirm it can read your page. Personal is $48 a month and Professional $87 a month. A task can run 5 minutes max on Free and 60 minutes on paid plans, and data is kept for 90 days.

The downside: per-credit pricing gets expensive once you go past a few hundred pages. Negative reviews mention slow support, password resets that never arrive and charges after cancelling. The vendor's claim that robots adapt to site changes on their own is worth checking on your own pages.

Tip: point Browse AI at the pages that matter (a competitor's 50 best sellers), not a whole catalog. For 10,000 pages, a simple per-request API is cheaper.

Zyte API works as a single call: send the URL, and Zyte picks the proxy, the browser and the unblocking strategy for that site.

Its pricing is the most honest on one point. You pay per 1,000 successful responses, across five difficulty tiers. A plain HTTP request is $0.13 per 1,000 on pay-as-you-go, an advanced one $1.27. With browser rendering, it's $1.01 to $16.08 per 1,000. Monthly commitments of $100, $200 or $500 bring those down, as low as $0.06 and $7.68.

In practice, 10,000 easy pages cost $1.30, and 10,000 heavily protected pages with a browser about $160.80. That 1-to-100+ spread shows the target site sets the price far more than the tool does.

The limit: you need a developer, and the tiered grid takes a few trial runs to see which tier your site lands in. Tip: spend the free $5 on a few hundred pages, check the tier you were billed, then size your monthly budget.