What is a scraping credit, and why can one request cost 75 of them?

A scraping credit is a vendor-defined billing token, not a standard amount of traffic, compute, or money. One request can cost 75 credits when a vendor assigns an expensive mode, such as stealth access or ultra-premium routing with JavaScript rendering, 75 tokens instead of the 1 token charged for a basic request. (ScrapingAnt, ScraperAPI, and ScrapingBee, retrieved July 26, 2026.)

What vendors mean by “credit”

There is no industry definition. The useful definition is the one in the vendor’s billing documentation:

What vendors mean by “credit” table

That last row shows why credit totals cannot be compared directly. Hyperbrowser publishes a fixed dollar value, while ScraperAPI and ScrapingBee sell monthly allowances whose request yield changes with the mode selected. Diffbot uses the same word for several product activities. “One million credits” is therefore not a useful comparison until you convert it into the exact work you need. (Hyperbrowser, ScraperAPI, ScrapingBee, and Diffbot, retrieved July 26, 2026.)

Which features multiply the request cost?

The recurring cost drivers are JavaScript rendering, residential or premium proxies, stealth or anti-bot modes, AI extraction, difficult target domains, response size, and browser time. The table includes every published request-cost mode for the comparable scraping APIs covered here. A number without a plus sign is the total bundle cost. A number with a plus sign is added to the base.

Which features multiply the request cost? table

Several products call their allowance “credits” even though a single fetch is not their only meter:

Which features multiply the request cost? table

Why exactly 75?

At ScraperAPI, the published total for ultra_premium=true with render=true is 75 credits. At ScrapingBee, a stealth-proxy HTML request is 75 credits and currently includes JavaScript rendering. These are vendor bundle weights, not evidence that either service sent 75 billable requests to the target. (ScraperAPI and ScrapingBee, retrieved July 26, 2026.)

The same URL can therefore look like a 1-credit job in a basic datacenter mode and a 75-credit job after a protected-site mode is selected. It can also cost 125 credits at ScrapingAnt with residential access and JavaScript, 30 at Scrapfly with residential access and rendering, or 25 at ZenRows with rendering and premium proxies. The numbers describe each vendor’s internal schedule, not a shared quantity. (ScrapingAnt, Scrapfly, and ZenRows, retrieved July 26, 2026.)

Worked example: will 10,000 credits cover 10,000 requests?

ScrapeGraphAI’s Starter plan is $20 per month and includes 10,000 credits. Its published billing rules charge 1 credit for a Markdown scrape, 2 for a screenshot, 5 for structured extraction, and an additional 5 when stealth is enabled. Assume a 10,000-request month with 6,000 Markdown scrapes, 2,000 screenshots, 1,500 structured extractions, and 500 Markdown scrapes with stealth. (ScrapeGraphAI pricing, retrieved July 26, 2026.)

Requests = 6,000 + 2,000 + 1,500 + 500 = 10,000

Markdown:              6,000 × 1       =  6,000 credits
Screenshots:           2,000 × 2       =  4,000 credits
Structured extraction: 1,500 × 5       =  7,500 credits
Markdown plus stealth:   500 × (1 + 5) =  3,000 credits
                                             ------------
Total                                      20,500 credits

The request count matches the allowance, but the credit total does not. The workload averages 2.05 credits per request, so the 10,000-credit allowance covers 48.78% of the stated mix and leaves a 10,500-credit shortfall. (ScrapeGraphAI pricing, retrieved July 26, 2026.)

20,500 credits ÷ 10,000 requests = 2.05 credits per request
10,000 included ÷ 20,500 required × 100 = 48.78%
20,500 required − 10,000 included = 10,500-credit shortfall

The same pricing page lists one-time top-ups of 1,000 credits for $5 and 10,000 for $40. One pack of each supplies 11,000 credits, covers the 10,500-credit shortfall, and leaves 500 credits. Together with the $20 Starter subscription, the stated month’s purchase is $65. The packs stack with subscription credits and do not expire. (ScrapeGraphAI pricing, retrieved July 26, 2026.)

10,000-credit pack + 1,000-credit pack = 11,000 top-up credits
11,000 purchased − 10,500 shortfall = 500 credits left
$20 subscription + $40 pack + $5 pack = $65

This is plan arithmetic, not a success-rate forecast. The public pricing page does not state whether failed scrapes, blocked pages, or empty extraction results consume credits, so that uncertainty must remain explicit in a real forecast. (ScrapeGraphAI pricing, retrieved July 26, 2026.)

How to calculate your own consumption before buying

1. Split the workload into request classes. At minimum, separate static pages, JS pages, premium or residential access, protected-site modes, named domains, AI extraction, browser sessions, and large downloads. ScraperAPI, Scrapfly, and ScrapingAnt publish different costs across those request classes; browser sessions use a separate resource-metered model covered in the cloud-browser billing guide. (ScraperAPI, Scrapfly, ScrapingAnt, and Browserless, retrieved July 26, 2026.)
2. Use the vendor’s combination rule. ZenRows publishes replacement totals of 1, 5, 10, and 25. Scrapfly uses either a 1-credit datacenter or 25-credit residential base, then adds 5 for rendering. ScrapeOps adds 25 for LLM extraction on top of its selected proxy and rendering bundle. (ZenRows, Scrapfly, and ScrapeOps, retrieved July 26, 2026.)
3. Multiply each class, then sum. Use total credits = sum(requests in class × credits per request) + applicable add-ons. This follows the additive schedules published by Scrapfly, Hyperbrowser, and Firecrawl. Browser-session resource arithmetic stays in the cloud-browser billing guide. (Scrapfly, Hyperbrowser, Firecrawl, and Browserless, retrieved July 26, 2026.)
4. Apply the billable-response rule. ScraperAPI charges HTTP 200 and 404, ScrapingBee charges 200, 404, and 410, and Scrape.do charges 2xx, 400, 404, and 410. A “pay only for success” label is not a complete definition. (ScraperAPI, ScrapingBee, and Scrape.do, retrieved July 26, 2026.)
5. Verify one sample request before scaling. ScraperAPI provides a URL-cost endpoint, ScrapingBee shows cost in its request builder, Scrape.do returns Scrape.do-Request-Cost, Scrapfly returns X-Scrapfly-Api-Cost, and ScrapingAnt returns Ant-credits-cost. These observed costs are safer than assuming that a target stayed in the cheapest class. (ScraperAPI, ScrapingBee, Scrape.do, Scrapfly, and ScrapingAnt, retrieved July 26, 2026.)
6. Add headroom for mix changes, not an invented success rate. If the share of protected pages is unknown, calculate separate low and high cases from the published classes. Do not turn an undocumented assumption into a single precise request count.

Which vendors do not use credits?

Many credible alternatives expose the underlying meter instead of an abstract credit allowance. Their units are still not automatically comparable, but the billing unit is visible:

Which vendors do not use credits? table

The buying rule is simple: compare the cost of your workload, not the size of the allowance printed on the plan card. Convert every candidate into billable successful requests or results for your actual target mix, then add any rendering, proxy, extraction, bandwidth, browser-time, and overage meters the vendor publishes.