Per-GB Pricing

Per-GB pricing is bandwidth-metered billing, the dominant model for residential and mobile proxy supply. You buy gigabytes and the meter runs on traffic through the proxy rather than on requests completed. Decodo lists residential from $3.75/GB on a 3 GB plan down to $2.75/GB at 100 GB, with pay-as-you-go at $4.0/GB (decodo.com/pricing, read 2026-07-27). Webshare lists rotating residential from $3.50/GB at 1 GB down to $1.40/GB at 3,000 GB (webshare.io/pricing, read 2026-07-27). The shape is consistent across the category: the rate falls with committed volume, and the commitment is what you are being asked to forecast.

What counts against the meter is broader than most teams expect. Response bodies count. So do images, fonts, CSS and third-party scripts if you do not block them. Request headers, cookies and TLS handshake bytes count. Redirects count. Retries count. Failed requests count, including the ones that return a challenge page instead of your data, which means a bad week against a hardened target shows up on the invoice rather than in the dataset. Most vendors meter both directions, so large POST bodies count too.

This is why a headless browser is so much more expensive than a plain HTTP fetch of the same URL. The document alone is commonly 50 to 150 KB. A full browser page load with images, webfonts, analytics and ad scripts routinely runs into the megabytes, and you are paying for every byte of it to render a price you could have parsed from the HTML. Blocking non-document resource types through request interception is usually the single largest lever available on a proxy bill, and it is a few lines of configuration in Playwright or Puppeteer.

To compare models, convert everything to cost per successful record on your own page-weight profile. Run a sample crawl of a thousand pages, measure actual bytes at the proxy rather than estimating, and divide by the records you kept. Per-GB favors light document-only fetches at high success rates. Per-successful-request models favor heavy pages and low success rates, because failures and retries are the vendor's problem. Credit models sit in between and need the vendor's own estimator to price honestly. Do that arithmetic before signing a committed-volume contract, because the discount tiers are exactly where an overestimate gets locked in.