How Much Does Web Scraping Cost in 2026? A Worked Estimation Model

At published list prices (July 2026), web scraping costs anywhere from $7.50-$81.25 a month for a 30,000-page side project to $125-$2,500 for a 500,000-page price-intelligence pipeline — each range running from the cheapest documented route to the costliest listed one — and into five figures plus negotiation for multi-terabyte AI data collection. The spread is that wide because the bill hangs on three numbers most estimates skip: pages per month, bytes per successful page, and how many attempts each success costs you. This guide gives you the formula — pages x page weight x retry factor, converted to gigabytes — and walks three personas through it against documented per-GB proxy rates ($2.00-$7.35/GB) and per-request API rates ($0.25-$1.50 per 1K).

The five numbers that drive your scraping bill

Every scraping budget is the product of the same handful of drivers. Three of them are properties of your workload that only you can measure; two are published prices you can look up. The table separates our modeling assumptions from documented rates.

Sources: Oxylabs residential pricing and Web Scraper API, Bright Data residential pricing and Web Unlocker pricing, Decodo residential pricing and Web Scraping API, IPRoyal residential pricing and Web Scraping API (as displayed/fetched July 17, 2026).

Engineer’s take (Hinata): The retry factor is where these estimates die — pages and prices are knowable up front, but nobody knows their block rate until real traffic flows, and on a per-GB plan every failed attempt still bills bandwidth. Before committing to any tier, I would run 100 representative URLs through a pay-as-you-go gigabyte and measure bytes per successful page, retries included; that single number decides whether per-GB or per-request wins your workload. If it lands anywhere near the 0.14-0.43 MB break-even band below, price both routes before signing anything.

The formula

Here is the whole model. It uses decimal units (1 GB = 1,000 MB), which matches how the arithmetic in provider tiers works out.

1. Traffic: monthly GB = pages per month x average page weight (MB) x retry factor ÷ 1,000. The retry factor sits inside the bandwidth term because failed and retried attempts transfer bytes you pay for on per-GB plans.
2. Proxy route: monthly cost = monthly GB x the per-GB rate of the tier your volume lands in. Full plan-by-plan tiers are in our residential proxy pricing comparison.
3. API route: monthly cost = pages per month ÷ 1,000 x the per-1K request rate. No retry factor where success-only billing is documented: Oxylabs' Web Scraper API, Bright Data's Web Unlocker, and IPRoyal's Web Unblocker bill only successful results; the failed-request policies of the Decodo and IPRoyal Web Scraping APIs are unverified.
4. Break-even page weight (MB) = per-1K API rate ÷ (retry factor x per-GB rate).

That last line is the decision rule. At a $2.50/GB proxy rate with a 1.4x retry factor, the break-even against Decodo's $0.50-per-1K standard API rate is 0.50 ÷ 3.50 ≈ 0.14 MB, and against the $1.50-per-1K rates (Bright Data Web Unlocker pay-as-you-go, Decodo premium with JS rendering) it is 1.50 ÷ 3.50 ≈ 0.43 MB. Lean JSON-like responses under 0.14 MB favor raw per-GB bandwidth; anything heavier than about 0.43 MB per successful page favors per-request pricing at those rates. Now the personas.

Persona 1: the side-project SEO tracker (30,000 pages a month)

Our modeling assumptions: 30,000 pages a month, 0.4 MB per page of lean rank-tracking HTML, and a 1.3x retry factor for lightly defended targets. Traffic: 30,000 x 0.4 x 1.3 ÷ 1,000 = 15.6 GB a month.

The verdict at this scale is not close: every documented per-request route undercuts the cheapest per-GB route, and the gap runs from roughly 1.7x to 8x. That follows directly from the break-even math — 0.4 MB x 1.3 = 0.52 MB billed per successful page, well above the 0.14-0.43 MB band. The subscription trap makes it worse: 15.6 GB does not fit Decodo's $35 10 GB plan, so the proxy route means either pay-as-you-go or paying $81.25 for a 25 GB tier you use 62% of. A side project also rarely has spare hours for retry logic, which tilts the field further toward APIs.

Persona 2: mid-size price intelligence (500,000 pages a month)

Our modeling assumptions: 500,000 product pages a month at 0.6 MB each — e-commerce pages with some client-side rendering — and a 1.4x retry factor for moderately defended targets. Traffic: 500,000 x 0.6 x 1.4 ÷ 1,000 = 420 GB a month. At the $2.50-$3.00/GB rates mid-size volume tiers land in (Bright Data's 332 GB tier is $3.00/GB; its 798 GB tier and the entry of Decodo's enterprise range are $2.50/GB), that is $1,050-$1,260 a month. Here are the actually purchasable routes:

On these assumptions the API column still wins — but this is exactly the band where the answer flips with page weight, so treat both tables as live options. Cut the average page to 0.2 MB (blocking images and media, hitting lighter endpoints) and traffic falls to 140 GB: $560 at Decodo's $4.00/GB pay-as-you-go, inside the API quotes' range. Push it to 1.0 MB and traffic hits 700 GB, about $1,750 even at $2.50/GB. Mixed difficulty moves the API column the same way: if half your targets need Decodo's premium routing, the blended bill is $500 (250K x $0.50 plus 250K x $1.50, per 1K). Measure first, then pick — and if per-GB wins, our cheapest residential proxies ranking shows where each tier boundary sits.

Persona 3: the AI training data pipeline (5 million pages a month)

Our modeling assumptions: 5 million pages a month at 1.2 MB each — article and media-heavy pages kept for training corpora — and a 1.4x retry factor. Traffic: 5,000,000 x 1.2 x 1.4 ÷ 1,000 = 8,400 GB, or 8.4 TB a month. That is beyond every published tier: the largest listed plans are Oxylabs' 1 TB Corporate at $2,500 ($2.50/GB, top-ups to 2 TB at unpublished rates) and Decodo's enterprise range topping out at 1,000 GB ($2.00/GB). Extrapolating those best published rates to 8.4 TB gives $16,800-$21,000 a month — an anchor for negotiation, not a quote. IPRoyal's marketing cites bulk rates down to $1.75/GB (vendor marketing claim), which would put the same volume at $14,700.

The per-request column decouples cost from bytes, which is exactly what you want when pages are heavy: 5M requests a month costs about $6,501 on Bright Data's Web Unlocker Scale plan ($499 for the included 383K plus 4,617K x $1.30 per 1K) or $7,500 at Decodo's $1.50-per-1K premium-plus-JS rate — roughly half to a third of the per-GB extrapolation. If a large share of your corpus is easy targets, the standard rates ($0.25-$1.00 per 1K, "from" pricing) pull that down further. At this scale everything is negotiable, and Proxyway's Proxy Market Research 2026 argues against waiting for better list prices: residential pricing has stabilized after declines of up to 75% across 2023-2025, and Bright Data is cited at roughly $300M annualized revenue growing 50% year over year on AI demand. Get written quotes from at least two providers and let them bid; our guide to proxies for AI training data covers the sourcing and compliance side of these pipelines.

The hidden costs the formula misses

1. Engineering time for retry and unblock logic. Per-GB bandwidth is a raw material: you build and maintain the retry policy, session rotation, failure classification, and target-specific unblocking yourself, and you keep fixing it as targets change. Per-request products bundle that work into the rate. We cannot put a documented dollar figure on your engineering hours, but for small teams this line item routinely decides the build-versus-buy question before the bandwidth math does.
2. Failed-request billing asymmetry. Oxylabs' Web Scraper API charges only for successful results, Bright Data's Web Unlocker bills only successful requests, and IPRoyal documents per-successful-request billing in its help center for its Web Unblocker specifically. Decodo's Web Scraping API and IPRoyal's Web Scraping API both list per-request pricing without an explicit failed-request policy on the pages we fetched July 17, 2026 — unverified, so ask before assuming. On per-GB plans, every failed attempt bills its bytes; that is the retry factor working against you.
3. VAT. Decodo lists every price excluding VAT: the $81.25 plan from Persona 1 becomes $97.50 for a buyer with 20% VAT. Oxylabs, Bright Data, and IPRoyal show USD prices without a VAT annotation, but your invoice may still add tax depending on billing country.
4. Promotional fragility. Bright Data's $4.00/GB pay-as-you-go rate is displayed as a 50%-off promo from an $8.00/GB list price, and Oxylabs' Web Unblocker page carried a 40%-off-for-six-months code (WU40) when we checked. Budget at list rates and treat the discount as upside — Proxyway's 2026 research notes Oxylabs and IPRoyal already retired long-running discount codes in favor of permanent plans revised down about 25%.

When to use a scraping API instead of proxies

The short version: buy per-request when your pages are heavier than the ~0.43 MB break-even, when targets need JS rendering or active unblocking, or when you have no engineering budget for retry infrastructure — and buy per-GB when pages are lean, volume is high, and you already operate a crawler. There is also a middle ground worth knowing about: per-GB unblockers, which bundle unblocking into bandwidth pricing — Oxylabs Web Unblocker runs $7.50-$9.40/GB and Decodo's Site Unblocker $6.75-$10.00/GB plus VAT, versus the per-request unblockers from Bright Data ($1.50 per 1K, 5,000 free monthly) and IPRoyal ($0.70-$1.00 per 1K). We compare the two product shapes head-to-head in scraping API vs proxy and rank the unblocker products in our web unblocker comparison.

Our methodology

This is a document-based estimation model, not a benchmark. Every dollar figure comes from the four providers' published pricing pages and help-center documents fetched on July 17, 2026, with vendor performance and marketing figures labeled as claims. The page weights (0.4, 0.6, and 1.2 MB) and retry factors (1.3x-1.4x) are our modeling assumptions, chosen to make the arithmetic concrete — substitute your own measurements before making a purchasing decision. ProxyFacts has not yet run first-hand benchmarks; every figure above comes from the cited provider pages and dated third-party research. ProxyFacts earns disclosed affiliate commissions when readers purchase through the CTA buttons in this article, and links to providers carry sponsored attributes; commissions never change the numbers we publish.

Frequently asked questions

What retry factor should I plug into my own estimate? No provider publishes one, so measure it: run 100 representative URLs and divide total requests by successful pages. Our 1.3x-1.4x band is a modeling assumption for lightly-to-moderately defended targets; hard targets can run worse, which is precisely when success-billed per-request pricing protects you.

Can I test my cost model without spending anything? On the API side, yes: Decodo's free plan includes 2,000 requests a month, Bright Data's Web Unlocker includes 5,000 free requests monthly with no card, and Oxylabs' Web Scraper API trial covers up to 2,000 results. On the per-GB side, Decodo offers a 3-day trial, Oxylabs a once-per-client trial on request, and IPRoyal has no standard trial — its FAQ points to the $7.35 1 GB minimum purchase.

Why did my real bill come in above the estimate? Almost always page weight or retries: unblocked images and media, heavier JS payloads, or a block rate above the assumed 1.3x-1.4x all inflate the bandwidth term. Promo lapses matter too — Bright Data's $4.00/GB is a 50%-off promotional rate — and Decodo invoices add VAT on top of every listed price.

Does per-GB always win at very high volume? No. At our Persona 3 assumptions (1.2 MB pages), 5 million pages a month costs $16,800-$21,000 extrapolated from the best published per-GB rates ($14,700 if you credit IPRoyal's marketing-cited $1.75/GB bulk rate) but $6,501-$7,500 at documented $1.30-$1.50 per-1K rates, because per-request pricing ignores page weight. Above the published tiers, both routes end in a negotiated contract anyway.

For the full playbook — tooling choices, block avoidance, and compliance — read our complete web scraping guide.