What per-GB proxies really cost a scraping pipeline

Scraping cost is a function, not a price. Pages per day, bytes per page, retry rate and how often the job runs all feed into the monthly number, and per-GB billing multiplies every one of those inputs by a rate. This page models that function for per-GB mobile residential proxies and for a WeProx line, which is one dedicated US mobile device at a flat price with unlimited data. You get the break-even formula, a cost-per-thousand-pages table, a worked nightly pipeline, and the conditions under which per-GB is the correct engineering choice.

The cost model

For a per-GB provider, monthly cost is roughly pages per month, times average bytes per page, times one plus the retry ratio, times the price per GB. For a flat line it is the number of lines times the monthly price, and bytes do not appear in the equation at all. The variable that decides the outcome is therefore not the price list but your payload size.

Rates used here: a public US per-GB mobile residential list from September 2026, at $20 per GB pay-as-you-go, $40 for 2 GB, and $300 for 25 GB, which is $12 per GB. On our side, a 4G line is $60 a month and a 5G line $75, both with unlimited data. Break-even is monthly price divided by per-GB rate: $60 over $20 is 3 GB, and $60 over $12 is 5 GB. For 5G, $75 gives 3.75 GB and 6.25 GB. Decimal units throughout, so 1 GB means 1,000 MB.

Note what the per-GB model is sensitive to. Doubling payload doubles cost. Adding a retry loop adds cost. Switching from raw HTTP to a headless browser can multiply it. None of those changes touch the flat-line figure.

Cost per 1,000 fetches by payload size

Payload size is where most estimates go wrong. Fetching raw HTML or a JSON endpoint is light; rendering a full page in a headless browser with images, fonts and scripts is many times heavier. The table prices 1,000 fetches at three illustrative payload sizes.

The last column is the practical takeaway. A headless crawl reaches the price of a whole WeProx 4G month after about 1,500 renders, which is a small job. An HTML-only crawler has far more room, but 20,000 fetches a month is still only around 670 a day.

Worked example: nightly price monitoring

Consider a retail analytics pipeline that checks 30,000 product pages every night from US mobile IPs, parsing price and stock status. The crawler requests HTML only, averaging 150 KB per page, and retries about one request in ten.

Nightly volume: 30,000 pages at 150 KB is 4.5 GB, and retries bring it to about 4.95 GB. Monthly, that is roughly 148 GB. At the $12 package rate the per-GB cost is about $1,780, which in practice means buying six 25 GB packages for $1,800. At pay-as-you-go it would be close to $2,970.

On WeProx the constraint is throughput and parallelism, not bytes. A 4G line typically sustains 20 to 45 Mbps; even at 20 Mbps, 5 GB moves in a little over half an hour of transfer time, so bandwidth is not the bottleneck. Request rate is. The team runs two 4G lines, one in Houston and one in Chicago, splits the URL list between them, rotates per request through the API, and pays $120 a month. Page counts can grow substantially without changing that figure; only a need for more concurrency would justify a third line. Keep crawl rates reasonable and respect each target site's terms, since a proxy changes the network path, not the rules.

When per-GB is the right engineering choice

Flat pricing is not a universal win, and the model above shows exactly where it loses. The conditions below favor a per-GB pool.

Many teams run both: a pool for the long tail of countries and occasional probes, and a set of flat lines for the high-volume US workload.

• Low volume: under 3 GB a month, for example a daily check of a few hundred HTML pages, the per-GB bill is smaller.
• Geographic breadth: WeProx lines are in eight US metros. Crawls that need exits in many countries belong on a global pool.
• Extreme fan-out: if you need thousands of distinct source IPs concurrently, a single device holding one address at a time is the wrong primitive.
• One-off probes: a quick test of a few hundred megabytes that will never run again costs a few dollars on pay-as-you-go. For anything bigger, even a one-off, a 4G day at $4 or a week at $26 usually comes out lower.

Variables that move the per-GB bill

If you keep per-GB for part of the stack, a few engineering choices change the invoice more than the rate does. Block images, fonts and media in headless browsers when you only need text. Prefer JSON endpoints over rendered pages where a site offers them. Send Accept-Encoding so responses arrive compressed. Bound your retries and log them, since most per-GB plans bill any bytes that cross the gateway, including responses that turn out to be blocks or errors.

On a WeProx line those optimizations still make a crawl faster and lighter on the target, but they stop being a billing concern. That changes design decisions: you can render fully when fidelity matters, and validate against real mobile page weight without watching a meter.

Line specs that matter to a pipeline

Each WeProx line is one physical 4G or 5G modem, router or Android phone with an AT&T, T-Mobile or Verizon SIM, dedicated to one customer. Endpoints are HTTP(S) and SOCKS5, with host, port, username and password shown in the dashboard, and UDP is available over SOCKS5. Rotation is unlimited, triggered by a link, the API or per-request mode, with no delay between changes. Addresses come from the carrier's regional CGNAT pool, so repeats within a day are expected and should not be treated as a fault. Because those addresses are shared by many real phones, target sites tend to evaluate mobile traffic on behavior, another reason to keep request patterns sane.

Lines can be moved between New York, Los Angeles, Chicago, Houston, Phoenix, Miami, North Carolina and Boston at no cost, with Boston running 4G only. Payment by card, PayPal, Zelle, wire or crypto; support 24/7 by email and Telegram bot.

Questions about per-GB pricing