Guide
Proxy Pricing Explained: Models, Multipliers and Real Cost
Proxy pricing is more than a per-GB number. Compare bandwidth, IP quality, targeting, rotation, and support to choose a plan that fits your workload.
Comparing proxy providers on headline price is close to meaningless, because the headline number measures different things at different providers and omits the terms that determine what you actually pay. A $3/GB service can cost less per usable record than a $1/GB one, and a plan with a low rate can cost more in total than a higher-rate plan with better terms. This explains the pricing models, the terms that distort them, and how to build a comparison that reflects reality.
The Pricing Models
Per gigabyte. The dominant model for residential proxies. You pay for data transferred, and the IP pool is shared infrastructure you draw from. Suits variable workloads and makes cost scale with output.
Per IP, per month. Standard for dedicated datacenter and ISP proxies. You rent specific addresses exclusively for a period, usually with unmetered or generously-capped bandwidth. Suits steady workloads needing persistent identity, and becomes very cheap per gigabyte at high utilization.
Per request. Used by scraping APIs and some gateway products. You pay per successful request regardless of page weight. Simple to budget, and the provider absorbs the retry risk — which is genuinely valuable, though you pay for it in the margin.
Subscription tiers with bundled allowances. A monthly fee including a bandwidth quota, often with features gated by tier. The gating is where the real cost frequently hides.
Hybrid. Many providers combine these — per-GB residential alongside per-IP datacenter on one account, which is usually the right structure for an operation with mixed workloads.
What Drives the Rate
Residential bandwidth costs more than datacenter for a structural reason: acquiring and maintaining consent-based residential IP supply is genuinely expensive, and the supply is finite. Datacenter addresses are cheap to provision.
Within residential, rate differences come from:
Pool quality. Fresh, low-reuse, widely-distributed addresses cost more to maintain than recycled ones. This is the difference that shows up as success rate.
Sourcing practices. Properly consented supply costs more than opaque supply. The cheap end of the market is frequently cheap for this reason, and the saving comes with legal exposure that transfers to you.
Geographic coverage. Depth in difficult markets is expensive to maintain.
Targeting granularity. City, ASN, and OS targeting require more pool metadata and more sophisticated routing.
A rate well below market usually indicates something about one of these, most often the first two.
The Terms That Distort the Rate
This is where comparisons go wrong. Check each one explicitly.
Bandwidth multipliers. Some providers charge the nominal rate for common markets and a multiple for others. An international program can find its effective rate is substantially above the advertised one. Ask for the full multiplier table, not a sample.
Concurrency tied to tier. The most consequential hidden cost. When throughput is an entitlement bundled with larger plans, reaching the parallelism your workload needs means buying bandwidth you will never consume. Your effective per-GB rate becomes the plan price divided by what you actually use, which can be several times the sticker price — and the overspend is invisible because the invoice shows the advertised rate.
This one is worth checking first, because it can invert an entire comparison.
Balance expiry. Bandwidth expiring monthly forces you to purchase for your peak month and forfeit the surplus in quieter ones. For bursty workloads — audits, project-based collection, corpus builds — this can mean paying for two or three times your actual consumption. Non-expiring balances let you buy at your average.
Overage pricing. Whether exceeding your allowance hard-stops or bills at a premium, and what that premium is.
Minimum commitments and contract length. Annual commitments at volume you have not validated are a common way to overpay.
Feature gating. City targeting, SOCKS5, API access, and session control sometimes sit behind higher tiers. If you need one, the tier it requires is your real price.
Support tiering. Response-time commitments by plan level matter if your operation is production-critical.
The Only Comparison That Matters
Headline rate tells you little. Compute cost per usable output:
cost_per_valid_record =
(bytes_per_attempt × attempts_per_valid × price_per_GB) / 1e9
Worked comparison on the same target:
| Provider A | Provider B | |
|---|---|---|
| Rate | $3.00/GB | $8.00/GB |
| Bytes per attempt | 180 KB | 180 KB |
| Valid rate | 68% | 96% |
| Attempts per valid | 1.47 | 1.04 |
| Cost per 1,000 records | $0.79 | $1.50 |
Here the cheaper provider genuinely is cheaper per record — the quality gap did not close a nearly 3× rate difference. But narrow the rate gap and the picture flips:
| Provider A | Provider C | |
|---|---|---|
| Rate | $3.00/GB | $3.60/GB |
| Valid rate | 68% | 96% |
| Attempts per valid | 1.47 | 1.04 |
| Cost per 1,000 records | $0.79 | $0.67 |
A 20% higher rate with a better success rate is 15% cheaper per record — plus the engineering time not spent diagnosing failures, which no formula captures.
The point is not that expensive is better. It is that rate alone cannot tell you, and the measurement takes an afternoon.
Costs That Are Not on the Invoice
Engineering time on failures. A provider producing frequent unexplained blocks consumes staff attention. At loaded cost, a few hours a week is material against most proxy bills.
Data quality. Silent failures — wrong-market geolocation, challenge pages recorded as valid, silent throttling producing partial renders — produce bad decisions. The cost of acting on wrong data is not proxy spend but it is real spend.
Lost collection windows. Where data has a time dimension — rank tracking, ad verification, fare pricing — a window missed is not recoverable. A provider whose throughput cannot complete a run inside the window imposes a cost that no per-GB comparison reflects.
Compliance exposure. Poorly-sourced residential supply carries legal risk that transfers to the customer. Increasingly a procurement question.
Migration cost. Switching providers means reconfiguring, re-verifying geo accuracy, re-tuning pacing, and re-measuring. Worth factoring into a decision that will be revisited in a year.
Building a Real Comparison
- Measure bytes per valid record per provider against your top three targets, 500+ requests each, counting parsed and validated records rather than HTTP 200s.
- Model monthly volume from your actual record counts and traffic mix.
- Apply the multiplier table to your real geographic distribution, not an average.
- Determine the tier you actually need — including for concurrency and any gated features.
- Model expiry waste against your consumption profile. For bursty usage, model peak-month purchasing.
- Add overage under a realistic over-consumption scenario.
- Compute annualized total, then divide by annual records for a true per-record cost.
Running that on the two or three providers on your shortlist takes an afternoon and routinely changes the answer relative to a price-page comparison.
Matching Model to Workload
| Workload | Best fit |
|---|---|
| Variable-volume collection | Per GB |
| Steady high-volume, permissive targets | Per IP datacenter |
| Persistent identity, account work | Per IP dedicated ISP |
| Bursty project work | Per GB, non-expiring |
| Low volume, retry risk unwanted | Per request |
| Mixed workloads | Hybrid on one account |
That last row is the common case and worth optimizing for. An operation routing permissive targets to datacenter and protected targets to residential pays substantially less than one applying residential to everything — and doing it on a single account avoids the overhead of managing two vendors.
FlameProxies prices residential bandwidth at $0.50/GB with non-expiring balances, offers residential, ISP, and datacenter on one account for per-workload routing, and applies no concurrency cap or request-rate limits — so there is no tier you have to buy into to unlock throughput, and the advertised rate is the effective rate rather than a starting point the terms adjust upward.