Web Scraping Pricing: What You Actually Pay (2026)

The same 10,000-page scraping job can cost $0 or $189 a month, and the tool is rarely the reason. The meter is. Web scraping pricing comes in four shapes: per-request API credits, prepaid platform usage, no-code task credits, and flat licences. Vendors pick the shape that makes their entry number look smallest, then leave the multipliers in the docs. The verdict up front: steady volume makes flat licences win, spiky volume makes credits win, and the multipliers below decide which one you are.

This page is the arithmetic the pricing pages skip. What a credit costs once JavaScript rendering or a stealth proxy multiplies it. Why monitoring burns credits while you sleep. Where a flat fee beats both. Every number was checked against vendor pricing pages or their own FAQs in October 2026, and the local-tool run below happened this morning.

The four web scraping pricing models

Model	How the meter counts	Entry price (Oct 2026)	Wins when
API credits (ScraperAPI, ScrapingBee, Firecrawl, Zyte)	Per request; JavaScript and protected sites multiply the cost	$0 free tiers; $16–49/mo paid	You call from code and volume is spiky
Prepaid platform usage (Apify)	Compute units, proxy GB, per-result actor fees	$0 ($5 usage/mo); $19/mo prepaid	You rent ready-made actors instead of writing code
No-code credits (Browse AI, ParseHub, Octoparse)	Rows, screenshots, pages per run, task slots	$0 capped tiers; $19–189/mo	Nobody on the team writes code
Flat licence or open source (Magpie, Screaming Frog, Scrapy)	Nothing per page	$0; $49/mo; £199/yr	Volume is steady and the machine is yours

One more tier sits above all four: managed scraping services with custom quotes, five figures a year for hands-off pipelines. That is a different purchase with a different sales cycle, covered in custom scraping quotes. Here we compare the meters you administer yourself.

API credits: one request, one to seventy-five credits

Every scraping API’s sticker reads “per request”. The fine print is that pages are not requests.

ScraperAPI’s free tier gives 1,000 credits a month that genuinely refresh, plus 5,000 bonus credits for the first week. One credit buys a basic HTML request. JavaScript rendering costs 5 to 10 credits, and a protected domain like Amazon or Google costs 25. Their $49 Hobby plan holds 100,000 credits, which is 100,000 easy pages or 4,000 Amazon pages. Read that sentence twice before picking a plan.

ScrapingBee prices the same way. The $49 Freelance plan advertises 250,000 credits; a stealth-proxy request burns 75, so the real capacity is about 3,300 protected pages. Their help center publishes the multiplier table, which is more honest than most. Zyte takes a different cut: pay-as-you-go at $0.13 per 1,000 simple HTTP responses, rising to $16.08 per 1,000 browser-rendered pages on hostile sites. That is a 123× spread between an easy page and a hard one, the lowest published entry rate I found among the major scraping APIs.

Firecrawl, built for LLM pipelines, bills 1 credit per page for markdown. JSON extraction adds 4 credits a page. A weekly crawl of a 100,000-page docs site is 400,000 credits a month in markdown, and over 2 million if you want structured JSON out of every page.

Apify is its own dialect of the same idea. Plans are prepaid usage: the $19 Starter plan buys $19 of platform credit, and unused credit expires at month’s end. Compute costs compute units (1 CU is 1 GB of RAM for 1 hour, $0.20 on Starter, down to $0.13 at Business), residential proxies add $8/GB on lower plans, and store actors with per-result pricing bill on top. The Apify alternative page works a full example: a Google Maps actor at roughly $4 per thousand places means 50,000 places a month costs about $200, every month, with no end to the meter.

What a web scraper price actually buys

So when a pricing page says “$49 for 100,000 pages”, read it as “for 100,000 easy pages”. The free tiers behind these APIs behave the same way: generous on plain HTML, gone in an afternoon once rendering or anti-bot proxies enter the job. The honest comparison is not plan price against plan price. It is (your pages × their multiplier) against the plan’s credit pool.

No-code credits: the meter runs while you sleep

No-code platforms replace “per request” with their own units, and monitoring is where the units leak.

Browse AI counts 1 credit per screenshot or per 10 extracted rows. Their own FAQ works the example: extracting 50 products from a listing page costs 5 credits, but monitoring 50 detail pages every 3 days burns about 500 credits a month, for one robot. That arithmetic is per-check, so it scales linearly: 500 product pages checked daily lands near 15,000 credits a month, 7.5 times the Personal plan’s 2,000. The free plan carries 50 credits a month; Personal carries 2,000 for $48 month-to-month or $19/month billed annually (12,000 credits a year). A monitored fleet eats that fast, which is the whole Browse AI alternative argument.

ParseHub meters by pages per run. The free plan stops at 200 pages per run, and Standard costs $189 a month as of October 2026, up from $149 in older reviews. The day your project needs 250 pages per run, the price jumps from $0 to $189. The ParseHub alternative page did the page math on this.

Octoparse sells seats instead. The free plan caps at 10 tasks and 50,000 exported rows a month; Standard runs $69/month billed annually. For steady high-volume work a seat is the cheapest of the three, but you pay it even in the months you barely log in.

Watch the expiry rules too, because vendors differ: Apify’s prepaid credit expires monthly, ScrapingBee’s 1,000 trial credits are one-time and do not refresh, and ScraperAPI’s free 1,000 come back every month. A “free” tier you can only spend once is a demo, not a plan.

Flat pricing: the meter disappears

Two flat models exist. The first is the yearly desktop licence: Screaming Frog’s crawler is free up to 500 URLs per crawl and £199 a year after, a price that has outlived several SaaS companies.

The second is local-first tooling. Magpie is the one I run daily, and its pricing has two halves. The core CLI is a free download under AGPL-3.0-or-later (or a commercial licence), with no run counter anywhere in it. The desktop app adds a GUI: 5 free runs per install, then $49 a month for one computer. A crawl costs 1 run whether it covers 12 pages or 12,000, failed and stopped runs are free, and there is no per-page meter at all.

$ magpie crawl 'https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html' \
    --corpus --max-pages 5 --max-depth 1 --out books.jsonl
crawl: run_id=5d2dd286c8c7bcdc-158435 pages_ok=4 pages_err=0 records=4

Four pages, 5.3 seconds, one run. On Firecrawl’s Hobby tier that is 4 of your 5,000 monthly credits. On Magpie’s counter it is 1 run, and in the CLI it would not even register.

Open source is the $0 end of flat: Scrapy and Colly charge nothing and scale as far as your code does. The honest ledger for that route is in the open-source web scraping tools comparison; the short version is a VPS at $40–80 a month, residential proxies at $4–8 per GB if the targets are picky, and your own hours. One 2026 total-cost-of-ownership breakdown puts self-hosted maintenance at 19–38 hours a month once anti-bot churn and breakage are counted, which at a blended developer rate is the most expensive line in this entire article.

Magpie’s limits belong in the same paragraph as its price, so: nothing runs when your machine is off (there is no hosted API, by design; the cloud vs local trade-off is exactly this), it will not solve CAPTCHAs, and the walled gardens (Google Maps, Instagram, LinkedIn behind login) are out of scope. After the 5 free runs, $49 a month flat is cheaper than a credit bill only if your volume is real. For a 50-page monthly job, free tiers win.

Price your own job before you pick a tool

Every pricing fight I have seen came from skipping this step. Work the numbers first:

monthly cost = plan fee
             + (pages × credits-per-page × price-per-credit)
             + proxy GB × $/GB
             + (maintenance hours × your hourly rate)

Job A: 500 product pages, price-checked daily, JavaScript-rendered. 15,000 requests a month. At ScraperAPI’s 5-credit render multiplier that is 75,000 credits, inside the $49 Hobby plan. Browse AI’s per-check arithmetic lands near 15,000 credits a month, as worked out above. On Magpie it is a nightly cron: $49 flat on the desktop licence or $0 on the CLI, because the licence buys unlimited runs, not credits.

Job B: a 100,000-page docs crawl, rerun weekly, feeding a RAG pipeline. Firecrawl in markdown is 400,000 credits a month, the $333 Growth plan; with JSON extraction it outgrows even the $599 Scale tier. Apify’s published estimates land around $100–200 a month of compute at that volume, plus proxy traffic. Magpie does it in 4 runs and 0 credits, but the meter that replaces money is time: about 28 hours per crawl at the default polite rate of one request per second, or a weeknight at 4 requests per second if the site tolerates it.

One line job A and B both miss: AI extraction. Schema-driven runs send your pages through an LLM on your own API key, and at 100,000 pages a month the token bill can pass the scraping bill. Point the provider at Ollama and the tokens run on your hardware for $0, which is the cheapest extraction available anywhere on this page.

Web scraping pricing FAQ

How much does web scraping cost per month?

From $0 to $999 and up. A small recurring job fits a $49 entry plan at most vendors. Around 100,000 pages a month runs $83–333 on hosted platforms. Enterprise and managed pipelines start at custom quotes. Self-hosted costs a VPS plus your maintenance hours.

Is there a truly free web scraper?

Yes, three kinds. Open-source tools (Scrapy, Colly, Magpie’s CLI) are free with no meter. Free tiers on hosted tools are real but capped: ParseHub at 200 pages per run, Octoparse at 10 tasks, ScraperAPI at 1,000 credits a month. And sites you control can be scraped by a script you wrote in an evening.

Do unused scraping credits roll over?

Rarely. Apify’s prepaid usage expires at month’s end, by their own pricing FAQ. ScrapingBee’s 1,000 trial credits are one-time. ScraperAPI’s free 1,000 credits refresh monthly. Check the rollover clause before the annual-plan discount looks attractive.

Why do JavaScript-heavy sites cost more to scrape?

Rendering a page in a headless browser costs the provider real compute, so they pass it on: 5–10× on ScraperAPI, 5× on ScrapingBee, and difficulty-based pricing that reaches 123× at Zyte. Protected domains add stealth-proxy multipliers on top, up to 75 credits a request.

The decision, in three lines

Under a few thousand pages a month, a free tier covers you; pick by output format, not price. Between 5,000 and 100,000 pages, multiply your page count by the render multiplier before you believe any plan’s headline, because that one number decides the tier. Past that, or at steady daily volume, flat licences and local machines win on unit cost and lose on wall-clock time, so pick the resource you are actually short of.

My own setup, after paying both kinds of bill: Magpie’s CLI on cron for the recurring jobs, a pay-as-you-go API key for the hostile sites it refuses by design, and no prepaid credit balance anywhere near my accounting.