Proxy pricing: per GB, per IP or per port for your job

Proxy pricing per GB, per IP and per port only compares once the job is fixed. How to size a job in GB, which unit fits it, and what else moves a bill.

Three billing units as three objects on one climb: a network switch with a module seated in each port held each month for datacenter, a cage of address modules with one module seated per static ISP address held each month, and a graphite GB package of gold data blocks, two wells already drawn down, metered by the GB for residential and mobile; in front, a glass slab sizes the job: 100,000 pages as HTML only, 2.2 GB, a thin sliver, against 100,000 full renders, a tall 216 GB prism
Quick summary · TL;DR
  1. Check the unit before the rate. Residential and mobile bill per GB, static ISP per IP per month, datacenter IPv4 per port per month, so a per-GB number next to a per-IP number tells you nothing on its own.
  2. Size the job in GB first. 100,000 pages a month come to about 2.2 GB as plain HTML and about 216 GB as full renders, and that gap picks the unit more than any rate card does.
  3. Small jobs are priced by the minimum unit. 1,000 HTML fetches move about 22 MB against a 1 GB smallest package, and proxymint sets no expiry on the GB, so the rest waits for the next run.
  4. You can skip the add-ons column on proxymint. City targeting, every rotation mode, HTTP and SOCKS5 all sit inside the same rate.

Proxy pricing comes in three units. Residential and mobile proxies bill per GB of traffic, static ISP proxies bill per IP per month, and datacenter proxies bill per port per month. On proxymint every tier is priced in volume bands, so the rate per unit falls as the order grows. Which unit fits depends on the job, not on the headline rate.

If you have two pricing pages open, one quoting per GB and one per IP, you have probably noticed they refuse to line up. They can’t, because a gigabyte and a month of one address are different goods. Neither number means much until the job is fixed, meaning how many pages, how heavy each page is and how strict the target is.

So size the job first and let the size pick the unit. Idle capacity and retries come last, and they are the easiest part of the bill to underestimate.

Proxy pricing comes in three units

Each tier sells in the unit that matches what it delivers. The types of proxies guide covers how the tiers differ beyond price.

Per GB. Residential and mobile proxies route traffic through a rotating pool of consumer and carrier connections. Bandwidth on those connections is what is scarce, so the meter counts bytes. How many IPs you touch and how many requests you send do not change the price; only the traffic does, which surprises anyone who expects the IP count to matter on a rotating pool. A package is one-off, with no renewal, the model behind pay as you go proxies.

Per IP per month. Static ISP proxies rent one ISP-registered address, assigned to your order for the whole billing term. The address is the product. It is not metered per GB, and it costs the same whether it carries one request or a million.

Per port per month. Datacenter IPv4 proxies rent a dedicated address on hosting infrastructure. Ports carry no bandwidth cap, so a busy port is billed exactly like an idle one.

Two consequences follow. On a per-GB meter, page weight drives the bill. With per-IP and per-port pricing, the number of addresses and months drives it, whatever the traffic.

Proxy prices on every proxymint tier are set in volume bands, with a lower rate per unit at higher quantity; the pricing page lists them. What other providers charge per unit, with sources, is in the monthly proxy pricing index. A first purchase comes with a 7-day money-back window under the terms.

One-off package or monthly term

Billing rhythm is the other half of the unit, and it is easy to miss when you only compare headline rates. Per-GB packages on proxymint are one-off: you buy one, it does not renew, and proxymint sets no expiry on the GB, so whatever a job leaves unused stays on the balance. Static ISP IPs and datacenter ports bill monthly per address, busy or not, because the address itself is what you rent.

Plans that bundle a monthly traffic allowance work differently. Before you put one next to a package, check whether unused traffic resets at the end of each cycle, because a plan that resets turns every quiet month into capacity you paid for and never touched.

Per result, a different product

Managed scraping APIs bill per successful result instead of per proxy unit. That vendor runs the retries, rendering and rotation, so the fair comparison is that per-result rate against a proxy bill plus the retry and rendering setup you would run yourself.

Size the job before comparing

In proxy pricing, a rate per GB means nothing until the job has a size in GB. Work it out as pages times bytes per page, plus headroom for retries.

Good starting figures come from the HTTP Archive Web Almanac 2025 page weight chapter (published 2026-01-15): the median mobile page transfers 2,164 KB, and the median mobile home page carries 22 KB of HTML. That is a gap of roughly 98x between fetching the HTML and rendering the full page with images, scripts and fonts.

HTML only. 1,000 pages at 22 KB is about 22 MB. 100,000 pages is about 2.2 GB.

Full render. 1,000 pages at 2,164 KB is about 2.2 GB. 100,000 pages is about 216 GB.

Real targets vary a lot, and a median is only a first guess. Measuring a sample takes ten minutes:

  1. Fetch 20 representative pages with the client the job will actually use.
  2. Sum the transfer sizes from its own logs or a HAR file, and divide by 20.
  3. Multiply that average by the page count.
  4. Add 20-30% for retries and redirects.

One trap in step 2: read wire bytes, not decompressed bytes. In Chrome DevTools that is the “transferred” figure at the bottom of the Network panel, not “resources”. In Python, len(response.content) from requests counts the body after it has been decompressed, so it overstates what a traffic meter sees whenever compression is on.

Asking for compressed responses with the Accept-Encoding header, defined in RFC 9110 section 12.5.3 (2022-06), keeps the HTML figure near the median instead of well above it.

Proxy pricing on three fixed jobs

With the job sized in GB, the proxy cost comparison comes down to one question: does the job spend more on bytes or on addresses? Traffic drives a per-GB bill; addresses and months drive a flat one.

Flat addresses suit steady traffic through a few exits on a target that accepts them. Light or bursty traffic spread over many exits suits per GB.

JobTrafficUnit that fits
1,000 HTML fetches, onceabout 22 MBPer GB (the 1 GB minimum covers it)
1,000 full renders, onceabout 2.2 GBPer GB for a one-off run
100,000 HTML fetches a monthabout 2.2 GBPer GB on strict targets
100,000 full renders a monthabout 216 GBPer port on lenient targets, per GB on strict ones
One endpoint every minute, 30 daysabout 0.4 GBPer GB
One endpoint every 10 seconds, 30 daysabout 2.6 GBOne port or one ISP IP

Job 1: 1,000 page fetches

At this size the minimum unit sets the price. 1,000 HTML fetches move about 22 MB, roughly 2% of the smallest 1 GB package, so the job starts with a full gigabyte on any per-GB tier. Since proxymint sets no expiry on GB, the rest of the package stays on the balance for the next run.

Full renders move about 2.2 GB, which needs a 3 GB package. Flat units run for a month either way, and a one-off job leaves most of that month idle. For a single run, per GB fits; for the same run every week, a flat unit becomes an option.

Job 2: 100,000 pages a month

This is where the units split apart. As HTML, the job is about 2.2 GB, well inside a small per-GB package. As full renders it is about 216 GB, almost a hundred times the traffic, and a per-GB bill follows it up.

Flat tiers need a number of addresses, not a number of bytes. Spreading 100,000 pages over 10 addresses is 10,000 each a month, about 14 an hour, and 10 ports are billed the same for HTML and for full renders. That makes flat ports economical for heavy pages on a lenient target. Where the target classifies hosting ranges, residential per GB is the unit that fits, and that call belongs before the job runs, not after the first wave of blocks.

Fixed ISP addresses serve a different job: holding one identity over weeks, which broad collection does not need.

Job 3: one endpoint, polled all month

Polling one endpoint is byte-light. Assume 10 KB on the wire per request, response plus headers, and check the real figure with a few requests. Once a minute for 30 days is 43,200 requests, about 0.4 GB. Every 10 seconds is 259,200 requests, about 2.6 GB.

Here per GB and flat pricing both work. Once a minute fits the smallest per-GB package with room to spare. At every 10 seconds the traffic is steady and comes from one place, and polling rarely needs a fresh address per request, so a flat unit fits well: a datacenter port if the endpoint accepts hosting ranges, an ISP IP if it expects one consistent ISP-registered address.

Hidden costs

A rate is only part of the total. Volume, retries and idle capacity change what a job really costs; location, rotation and protocol do not.

Volume bands

Proxy pricing on every tier drops per unit as the quantity grows, and checkout prices the whole package at the rate for its size. Size the package from the job estimate plus headroom, then read the rate for that quantity on the pricing page.

Retries and blocked requests

On a per-GB meter, a blocked request still spends bytes: the challenge page, the redirect, the retry. Say 70% of your requests return the content you wanted, each good page carries the traffic of about 1.4 requests (1 / 0.7), and the effective rate climbs with it. That is what the 20-30% headroom in the sizing step pays for.

Count success as the content you needed, not a status code. Soft blocks often answer 200 OK with a challenge page, which looks fine in the logs and still costs traffic. With flat pricing a failed request adds no bytes to the bill, but it burns the address’s time, and an address the target challenges on sight adds cost without adding pages.

Unused capacity

Every unit carries the same trap: paying for capacity the job never uses. An honest per-unit cost is the bill divided by what was actually used. Ten ports carrying 100,000 requests a month spread their bill over a hundred times as many requests as the same ports carrying 1,000. For scraping, divide by successful pages rather than requests sent. It works on any quote you already hold, which is handy when the quotes in front of you use different units.

What does not change the price

On proxymint, location does not change the rate: country, region and city targeting on the per-GB tiers bill the same. Neither does rotation, which is set per order as a new IP on every request, a timer from 5 to 60 minutes, or sticky while the device stays online. HTTP and SOCKS5 are both included on every tier, and HTTPS sites run through either one via the CONNECT method described in RFC 9110 section 9.3.6.

Cheapest proxy type by target

The cheapest proxy type is the one the target accepts. A blocked request returns no data at any rate, so proxy pricing per unit comes second.

Lenient target, heavy pages. Datacenter proxies fit, because a port is billed the same at 2 GB or 200 GB. Targets that classify hosting ASNs call for a consumer address instead, chosen before the run; datacenter vs residential proxies compares the two.

Strict target, light pages. Rotating residential proxies fit: consumer addresses, billed only for bytes moved. Fetching HTML or JSON instead of rendering keeps the bill small, and it is usually the first change worth making to a scraper.

One address that must hold. Static ISP proxies carry an ISP-registered address, assigned to your order for the whole billing term, billed per IP per month. Long logged-in sessions and partner firewalls that need one fixed IP belong here. How it differs from a datacenter port, which also holds one IP, is covered in ISP vs datacenter proxies.

Mobile origin required. Mobile proxies are billed per GB and exit from mobile carrier networks, classified as cellular by MaxMind, which app-first platforms and mobile ad checks expect. Price monitoring and catalog scraping usually clear on residential. What one carrier GB buys, and how to size a mobile job, is in mobile proxy pricing.

Units differ between tiers because supply differs. Datacenter addresses sit in hosting ranges run in bulk, so they are rented per address. Residential and mobile exits sit on home and carrier connections, where upstream bandwidth is limited, so their traffic is metered per GB.

Choosing a tier by the job

Proxy pricing only compares once the job is fixed, so fix it first: page count, bytes per page from a 20-page sample, and whether the target accepts hosting ranges.

  • If the target is lenient and pages are heavy, price datacenter ports; heavy, steady traffic sits well on a flat port.
  • If the target blocks hosting ranges, price residential per GB, and fetch HTML or JSON rather than rendering. 100,000 HTML fetches fit in 3 GB.
  • When one address must hold for weeks, price static ISP per IP per month.
  • If the platform expects a carrier origin, price mobile per GB for the part of the job that needs it.
  • Size one package per job from the estimate, and read the rate for that volume on the pricing page.

Frequently asked questions

It depends on the unit and the tier. On proxymint, residential and mobile are billed per GB in one-off packages, static ISP per IP per month and datacenter IPv4 per port per month. Every tier has a lower rate per unit at higher volume, and the current numbers are on the pricing page.

They run on different infrastructure. Datacenter addresses sit in hosting ranges run in bulk, so they are rented per address with no bandwidth cap. Residential exits sit on consumer connections with limited upstream bandwidth, so the traffic is metered per GB. Datacenter suits lenient targets and heavy pages; residential suits targets that screen hosting ranges.

Per GB fits byte-light work and jobs that need many different addresses, because the price follows traffic only. Per IP or per port fits heavy pages and steady traffic through a few addresses, because the price stays flat however many bytes move. Estimating the job in GB first shows which unit matches it.

The cheapest type is the one the target accepts, because a blocked request returns no data at any rate. On targets that accept hosting ranges, datacenter ports billed per port per month, with no bandwidth cap, carry heavy traffic on a flat bill. On targets that screen hosting ranges, rotating residential proxies billed per GB are the economical choice, picked before the job runs.

Hidden costs usually come from the job rather than the rate card: retries and blocked requests on a per-GB meter, idle addresses on a monthly plan, unused allowance on plans that reset each cycle, and paid add-ons for city targeting at some providers. On proxymint, location, rotation settings and protocol carry no surcharge, and no expiry is set on per-GB packages.

It depends on the provider and the plan, so look for the words expire or reset in the terms before you buy. On proxymint, per-GB packages are one-off and proxymint sets no expiry on the GB, so traffic a job leaves unused stays on the balance for the next run.

Per IP per month. An account that signs in from the same place every day needs one address that holds, and static ISP proxies give one ISP-registered address, assigned to the order for the whole billing term, billed per IP per month. Logged-in work moves little traffic, so the address count, one per login location, sets the bill. App-first accounts that expect a mobile carrier origin are the exception and run on mobile per GB.

Researched and drafted with AI tools, then checked against our publishing rules before it went live: 2 outside links, 12 links to other proxymint pages, no prices and a fact-check pass against our product records. How we write.