One wallet for proxies, browsers, and extraction

Run a scraper at any real volume and the cost shows up in three places. A proxy provider meters bandwidth by the gigabyte. A browser service bills by the minute a page stays open. A model API charges by the thousand tokens you send it. Three meters, three invoices, three dashboards to reconcile at month end, and not one of them tells you the per-page cost until after the job has already run.

webscrape.ai puts all three behind one credit balance. Proxies, browser time, and extraction come out of the same wallet, and the price of a successful call is a fixed number of credits you can multiply out before you start.

What each call costs

Web search is the one dynamic line: 1 credit per 5 results requested, rounded up, plus 1 credit if the stealth tier runs. Everything else is a flat cost per successful call, which is the point of the table. You can read your bill off it before you write a line of code.

A credit is charged on what a call returns, whatever it took to get there. A page that takes several fetch attempts and a megabyte of HTML costs the same as one that returns on the first try, because the credit is fixed per successful call and does not move with token usage. We absorb that variance, so your cost tracks how many pages you ask for and nothing else.

A page bills at base, or it bills nothing

Most pages never leave the cheapest fetch tier. The stack starts with a plain HTTP request and climbs to a stealthed browser only when a site forces it, and the stealth surcharge applies only on the pages where that tier actually ran. A request that clears on fast HTTP is billed at the base rate even if you left stealth available as a fallback. The mechanics of that escalation are in how the fetch tiers decide when to climb; the billing consequence is that you are not paying the heavy-tier price on the easy pages.

The other half of the floor is failure, and it is where the two cost models split hardest. A fetch that never returns usable data costs 0 credits here. Dead links, hard blocks, and timeouts do not reach the bill at all. On a metered stack those same failures still cost money: the proxy already moved the bytes and the browser already spent the minutes, and a 403 at the end of that refunds neither. You pay for the attempt; we charge for the result.

The math on a real job

Suppose you need 10,000 product pages as structured JSON. At 5 credits each, if every page clears the cheap tiers, that is:

10,000 pages × 5 credits = 50,000 credits

Some fraction will force the stealth tier. Only you know which sites you are hitting, so treat the split as an input you control. If 1 page in 10 needed it:

9,000 × 5  = 45,000
1,000 × 10 = 10,000
            ─────────
total       55,000 credits

Now the part the three-bill model hides. If 300 of those URLs are dead or block you outright, they fetch nothing and cost nothing, so they never enter the total. On a metered stack those 300 dead pulls would still have rung up bandwidth and browser time, because the meter runs on every request that goes out, including the ones that come back empty. Here the pages that returned are the only line on your bill, and you could read its size off the table above before the job ran.

If you are pinning a stable site, the floor drops further: save a working extraction as a Blueprint and the replay runs at 1 credit a page with no extraction step, the cheap deterministic path for as long as the page holds its shape.

When one wallet is not the cheaper answer

If you already run a proxy pool and a headless fleet at the scale where their marginal cost is near zero, a credit model competes with infrastructure you have already paid down, and the wallet's value to you is mostly consolidation rather than unit price. The reframe here is aimed at the team that would otherwise stand up and maintain those systems from scratch, where the real bill includes the engineer-weeks of keeping proxies rotating and browsers patched, not just the bandwidth.

Credits also abstract away the dollar figure on purpose. For what a credit costs and where the plan tiers sit, see the pricing page.

Pick the model you can predict

What you are really paying for is the ability to predict the bill before the job runs instead of reconciling three of them after. Grab a key and run a real extraction against a site that keeps breaking on you. A failed scrape costs nothing here, so the only pages you pay for are the ones that come back with data.

Get a key and run one.