Apify's plan prices are floors, not totals. Enter your catalogue size and refresh rate to see the compute units, proxy traffic and monthly figure behind the sticker price — and what the same volume costs here.
One page fetched per product, per refresh.
This one setting prices both columns: Apify gets 50 pages per compute unit on targets like these, and on our side about 25% of them need a browser render at €3.75 per 1,000, and the rest bill at €1.50.
Apify bills in dollars and we bill in euros, so one side is always converted — at a fixed 1.08, checked 21 September 2026, with the original figure shown underneath. Published rates stay in their own currency so you can check them against the source.
Cheaper than the tiers either side: the lower plan’s rates would overrun, the higher plan’s floor would sit idle.
Every page is billed at whatever that page needed, so a mixed catalogue lands between two published rates rather than on one of them. Proxy and rendering are already inside those rates — the number above is the number on the invoice.
Start freeThat's roughly 60% less than the Apify estimate above, both figures converted to USD at the fixed 1.08 rate. The percentage is deliberately coarse: the gap runs from about 1.25x to 21x depending on what your targets need, and a decimal place would imply a precision a month-old exchange rate does not have.
Every row assumes typical retail targets behind a residential proxy with a quarter of the catalogue needing a browser render — the same settings the calculator above loads with, at 50 pages per compute unit. The plan column is the cheapest Apify tier that survives the volume, picked the same way the calculator picks it.
| Products | Refresh | Pages/mo | Compute units | Proxy GB | Apify plan | Apify/mo | ScrapeWise/mo | You save |
|---|---|---|---|---|---|---|---|---|
| 1,000 | Weekly | 4,286 | 86 | 1.3 | Starter | $27 | $10 | $18 65% |
| 1,000 | Daily | 30,000 | 600 | 9.0 | Starter | $192 | $67 | $125 65% |
| 1,000 | Hourly | 720,000 | 14,400 | 216.0 | Business | $3,384 | $1,604† | $1,780 53% |
| 5,000 | Weekly | 21,429 | 429 | 6.4 | Starter | $137 | $48 | $89 65% |
| 5,000 | Daily | 150,000 | 3,000 | 45.0 | Scale | $818 | $334 | $483 59% |
| 5,000 | Hourly | 3,600,000 | 72,000 | 1,080.0 | Business | $16,920 | $8,019† | $8,901 53% |
| 25,000 | Weekly | 107,143 | 2,143 | 32.1 | Scale | $584 | $239 | $345 59% |
| 25,000 | Daily | 750,000 | 15,000 | 225.0 | Business | $3,525 | $1,671† | $1,854 53% |
| 25,000 | Hourly | 18,000,000 | 360,000 | 5,400.0 | Business | $84,600 | $40,095† | $44,505 53% |
| 100,000 | Weekly | 428,571 | 8,571 | 128.6 | Business | $2,014 | $955 | $1,060 53% |
| 100,000 | Daily | 3,000,000 | 60,000 | 900.0 | Business | $14,100 | $6,683† | $7,418 53% |
| 100,000 | Hourly | 72,000,000 | 1,440,000 | 21,600.0 | Business | $338,400 | $160,380† | $178,020 53% |
The savings column is simply the Apify figure minus ours, converted at the fixed 1.08 rate — subtract the two columns beside it and you get the same number. The percentage shrinks as volume grows and then flattens, because their compute rate improves in steps until it runs out of steps, while our page rate never changes.
† Past €1,000 a month, which is the largest self-serve top-up, so these rows are not a card-on-file purchase on either side. The arithmetic still holds — it just belongs in a managed service conversation rather than a signup form.
One CU is 1 GB of RAM running for one hour. A clean HTML page costs a fraction of one; a page behind a bot wall that needs a headless browser and three retries costs several times more. You are billed for time and memory, not for pages.
Billed per gigabyte, separately from compute. Residential IPs run $7–$8/GB depending on plan. At roughly 0.3 MB per product page, a 150,000-page month is 45 GB — often more than the compute line.
Third-party scrapers from the store carry their own monthly rental or per-result fee on top of the platform bill. This calculator assumes you wrote your own Actor, so it charges you nothing for this.
A compute unit is not a quantity of pages. It is 1 GB of allocated memory running for one hour — machine time, priced by the machine you reserved rather than the work you got out of it.
CU = memory in GB × runtime in hours
| Memory allocated | Runtime that burns 1 CU |
|---|---|
| 128 MB | 8 h |
| 256 MB | 4 h |
| 512 MB | 2 h |
| 1 GB | 1 h |
| 2 GB | 30 min |
| 4 GB | 15 min |
| 8 GB | 8 min |
Read that table twice, because it is where most of the surprise in an Apify invoice lives. The memory setting is a reservation, not a measurement. An Actor left on a 4 GB default that never touches more than 900 MB is spending compute units four times faster than one sized to fit — same code, same pages, same speed, four times the compute line.
It also explains why "pages per compute unit" varies so wildly. A static HTML page parsed at 512 MB in half a second and a JavaScript storefront rendered in a headless browser at 4 GB over eight seconds differ by a factor of well over a hundred in what they cost, even though both produce exactly one price.
That is the whole reason the difficulty toggle at the top of this page moves the total more than any other control. It is standing in for a variable — how much machine your targets force you to spend — that you cannot know until you have already run the job.
Rates read off apify.com/pricing on 21 September 2026. Every figure below feeds the calculator directly — there is no second set of numbers.
| Plan | Monthly floor | Included credit | Per compute unit | Residential proxy |
|---|---|---|---|---|
| Free | $0 | $5 | $0.20 | $8.00/GB |
| Starter | $19 | $19 | $0.20 | $8.00/GB |
| Scale | $199 | $199 | $0.16 | $7.50/GB |
| Business | $999 | $999 | $0.13 | $7.00/GB |
The middle setting. Clean static pages get 150; a hostile target needing a full browser and retries gets 20. The spread between the two is 7.5x, which is why the difficulty toggle moves the total more than anything else on the page.
A typical retail product page after images are skipped. Sites that force a browser render pull more, so treat the proxy line as a lower bound.
Flat. A 31-day month adds about 3% and a leap February takes 7% off, neither of which changes which plan you land on.
Every paid Apify plan includes platform credit equal to its monthly price. That makes the price a floor rather than a discount: spend less and you still pay the sticker, spend more and you pay what you spent.
Apify publishes $8 on the lower plans and $7 on Business. The $7.50 for Scale is the rate their own worked example uses and the only one that reconciles to the total they publish.
Plain €0.15, Render €0.75, Super €1.50 and Render + Super €3.75 per 1,000 pages. A unit test fails the build if this page and /pricing ever disagree.
5,000 products, refreshed once a day, on typical retail targets that need a residential proxy. That's 150,000 page fetches a month.
The same 150,000 pages here are €309.38, with no floor underneath them: a quarter of the catalogue on Render + Super at €3.75 per 1,000, the rest on Super at €1.50. That quarter isn't a number we chose independently — it follows from the same "typical" setting that gives Apify its 50 pages per compute unit. Move the calculator to clean targets and our month is €225 while their compute drops too; move it to hostile and ours is €478.13 while theirs passes $1,900. One assumption, applied to both columns.
The calculator at the top of the page already opens on these exact numbers, so this button only does anything once you've changed something — it puts the four inputs back and scrolls you up to them.
All four push the real bill up, never down. The figure above is a floor on a floor.
Store Actors carry monthly rental or per-result pricing on top of compute and proxy. A single rented Amazon or Zalando scraper can add $30–$100 a month before it has fetched anything.
Adds to the estimate.
Datacenter proxies are included up to a plan cap, then billed per IP. Google SERP traffic is priced separately again.
Adds to the estimate.
Committing to a year knocks a fixed percentage off the floor, but the metered overage above the credit is unchanged — and overage is what dominates at any real catalogue size.
Trims the floor only.
Lower plans cap how many Actor runs go at once, so a large daily refresh may not finish inside the day. The fix is either a higher plan or an engineer rewriting the scheduler.
Adds to the estimate.
The band above lists what this page doesn't count. This one is about the lines it does count reading low — each of these inflates compute or proxy against the same workload.
A page that returns a 403 and succeeds on the third attempt costs three page-loads of compute and three page-loads of proxy traffic. Nothing in the estimate above knows how often your targets refuse you, and the sites worth monitoring are the ones that refuse you most.
Compute and proxy, both.
The Actor memory setting is a reservation. An Actor left on the 4 GB default that only ever touches 900 MB burns compute units four times faster than one sized to fit, doing identical work at identical speed.
Up to 4x the compute line.
Hit the timeout and the run is killed, but the machine time up to that point is already metered. A misconfigured selector that hangs on every page can burn a day of compute and produce an empty dataset.
Full compute, zero output.
Reads, writes and key-value store operations each carry a unit price. Individually trivial; at a few million writes a month it becomes a line you notice, and it is not part of the compute or proxy figures above.
A third line, uncounted here.
None of this requires leaving the platform. In rough order of how much they save, on a typical price-monitoring workload.
Run the job once, read the peak memory off the run log, then set the allocation one step above it. This is the single largest lever on the page because compute is billed as memory × time — halving a 4 GB reservation that never needed it halves the compute line without touching the code.
Typically 50–75% off compute.
Datacenter proxies are included up to your plan cap; residential is $7–$8 per gigabyte on top. Fetch with datacenter, detect the block, and escalate only the URLs that actually got refused. On a catalogue where 80% of targets are indifferent, this removes 80% of the proxy bill.
Cuts the proxy line, not the compute.
Most catalogues have a small head that reprices constantly and a long tail that barely moves in a quarter. Refreshing the top 10% daily and the rest weekly produces roughly a third of the pages of a flat daily schedule — and loses almost nothing that matters commercially.
Around 70% fewer page fetches.
A great many retail pages ship the price in a JSON-LD block or a hydration payload in the initial HTML. Checking for that before reaching for a headless browser turns a multi-second render into a single fetch, and takes the target from the 50-pages-per-CU band towards 150.
Up to 3x more pages per CU.
An unbounded retry policy against a site that has decided to block you is a bill with no ceiling. Cap attempts, back off, and route persistent refusals to a separate low-frequency queue rather than letting them retry inside the main run.
Removes the worst tail cases.
The compute unit, the credit that is not a discount, and the lines most people forget to cost.
A compute unit is 1 GB of memory running for one hour. It is a measure of machine time, not of pages, so the same 10,000 products cost different amounts depending on whether the target serves static HTML or needs a headless browser and retries. On typical retail sites one CU buys roughly 50 pages; on clean static pages it buys about 150, and on a hostile target about 20.
Because the plan price is a floor, not a cap. Each paid plan includes platform credit equal to its monthly price — $19 on Starter, $199 on Scale, $999 on Business. Once your compute and proxy usage exceeds that credit you are billed for the excess on top of the floor. Residential proxy traffic at $7–$8 per gigabyte is the line that most often takes people past it.
Refreshed once a day on typical targets with a residential proxy, 5,000 products is 150,000 page fetches a month: about 3,000 compute units and 45 GB of traffic. On the Scale plan that is roughly $480 of compute and $337.50 of proxy, totalling about $818 a month. Drop the residential proxy and the same job is around $480; move to hostile targets and the compute alone is over $1,900.
No. It is a usage budget. Because the credit equals the plan price on every paid tier, your bill is whichever is larger — the sticker price or your actual usage. Spending $50 on a $199 plan still costs $199; spending $400 costs $400. Two intuitive readings are both wrong: the floor is not added to your usage, and the credit is not subtracted from a bill that still includes the floor.
Yes, and it is usually the part people forget to cost. At roughly 0.3 MB per product page, 150,000 pages a month is about 45 GB. At $7.50 per gigabyte that is $337.50 on top of compute — comparable to the compute line itself. Sites that force a full browser render pull considerably more than 0.3 MB per page, so the real figure tends to be higher.
It is accurate for the three things it models — compute units, residential proxy traffic and the credit-versus-floor arithmetic — using rates published by Apify on 21 September 2026. It deliberately excludes rented Actor fees, extra datacenter IPs, SERP proxy and engineering time, all of which push the real bill up. Treat the output as a lower bound, not a quote.
We bill per page fetched, at one rate per tier: €0.15 per 1,000 plain pages, €0.75 with browser rendering, €1.50 with a residential proxy and €3.75 for both together. Because each page is billed at whatever that page needed, a mixed catalogue lands between two of those rates rather than on one — the 150,000-page example above is €309.38 with a quarter of it rendered, €225 with none and €562.50 with all of it. There is no plan floor underneath any of those and no compute unit to convert. Five free requests are available before you pay anything.
Four levers, in descending order of size. Right-size the Actor memory, because compute is billed as memory × time and most Actors sit on a default reservation they never use. Try datacenter IPs first and escalate to residential only on the URLs that actually got blocked. Tier your refresh rate so the fast-moving head of the catalogue runs daily and the tail runs weekly. And check whether the price is already in the initial HTML before reaching for a headless browser. Together these routinely halve a bill without changing what you collect.
Allocated. The Actor memory setting is a reservation, and the compute unit is defined as 1 GB of that reservation running for one hour. An Actor left on a 4 GB default that only ever touches 900 MB burns compute four times faster than one sized to fit, at identical speed and identical output. It is the most common reason a real bill runs ahead of an estimate.
Yes, both. A page that is refused twice and succeeds on the third attempt bills three page-loads of compute and three page-loads of proxy traffic. A run that hits its timeout and gets killed still bills every second of machine time up to that point, even though the dataset is empty. Neither is modelled in the estimate on this page, which is one of several reasons to treat it as a lower bound.
No plan floor, no credit to burn through, no compute unit to convert. One rate per page fetched, and five free requests to prove it works on your targets before you pay anything.
