Per-GB networks: the formula#
monthly GB = requests per month × average bytes per request (both directions) ÷ 1,000,000,000, then read the rate for a purchase of that size on the ladder. The tier applies to the whole purchase, so buying the month at once costs less than buying it in pieces.
| Request type | Typical size | Notes |
|---|---|---|
| HTML product or listing page | 200–500 KB | Compression on; no assets loaded |
| JSON API response | 20–200 KB | The cheapest way to collect |
| Search results page | 200–800 KB | Heavier with scripts inline |
| Headless browser, assets blocked | 0.5–1.5 MB | HTML plus XHR calls |
| Headless browser, everything loaded | 2–6 MB | Images and fonts dominate: block them |
Worked examples on the residential ladder#
| Workload | GB per month | Rate | Monthly cost |
|---|---|---|---|
| 100,000 product pages at 350 KB | 35 | $1.27/GB | $44.59 |
| 500,000 pages at 350 KB | 175 | $1.08/GB | $188.65 |
| 2,000,000 API calls at 60 KB | 120 | $1.08/GB | $129.36 |
| 50,000 rendered pages, assets blocked, 1 MB | 50 | $1.23/GB | $61.25 |
| 50,000 rendered pages, everything loaded, 4 MB | 200 | $1.08/GB | $215.60 |
Rates are the current Proxy Cyber residential tiers; the pricing page has the full ladder and a calculator that uses the same table. Mobile traffic follows its own ladder, from $1.83 per GB.
Per-IP networks: the other arithmetic#
ISP and datacenter addresses are priced per IP and per term, so the question becomes: how many parallel identities or how many fixed exits do you need? One address per account, per whitelisted egress or per parallel task; multiply by the monthly rate of your term and zone; apply the volume discount from 10 addresses. Traffic is unlimited on ISP and pooled on datacenter (100 GB per address), so heavy pages cost nothing extra there.
Choosing by cost per successful request#
The cheapest network is the one whose requests succeed. A datacenter address at a fraction of a cent per request is expensive if the target blocks it; a residential gigabyte that returns real pages is cheap. Estimate success rates per network on a small sample (fifty requests), then divide cost by successes.
Five habits that halve the bill#
- Block images, media, fonts and trackers in browsers.
- Prefer JSON endpoints and mobile pages to full desktop HTML.
- Cache what does not change; re-fetch prices, not descriptions.
- Send datacenter-tolerant targets to datacenter addresses.
- Purchase traffic against a measured forecast, not only a lower unit price. Unused purchased traffic remains available, but a larger prepayment still ties up funds. Check the current tier and the terms applicable to your order.
Questions people ask#
Does traffic count in both directions?
Estimate traffic from a representative test, then compare the balance before and after it. Response bodies, uploads, browser assets and retries all affect consumption. Confirm the applicable metering and refund terms before paying; this guide does not promise that failed requests are free or that unused funds are automatically refundable.
What if I over-estimate?
Purchased residential and mobile traffic remains available until used. Static IPs and devices have an order term. Wallet refunds, if applicable, follow the confirmed terms and a manual review; keeping unused credit is not the same as a guaranteed cash refund.