Not All Proxy Pool Numbers Mean the Same Thing
Pool size is the first thing anyone checks when buying residential proxies, and rightly so. More addresses means a lower chance of reusing one the target site has already seen, which is the whole point of paying for a residential network in the first place.
The problem is not the number. The problem is that two providers can publish the same figure and be describing completely different things, and nothing on either website tells you which. Learning to read the number is worth more than memorising who claims the biggest one.
The same number, counted three ways
Cumulative over a window. Every unique address the network has seen in the last month, quarter, or since launch. Residential addresses churn constantly, because they belong to real connections that go offline, so a cumulative count keeps rising even if the network stops growing. It is a legitimate way to size a pool and it is what most published figures represent, including ours: 205M+ addresses across 195+ countries.
Live availability. How many addresses are reachable right now. Always a fraction of the cumulative figure, for every provider, because a residential network is a rolling window rather than a warehouse. Nobody publishes this as their headline because it moves by the hour and looks smaller.
Peak concurrent. How many distinct addresses the network can hand out simultaneously under load. This is the one closest to what a large scraping job actually consumes, and almost nobody reports it at all.
None of these is dishonest. They are answers to different questions, and a buyer comparing a cumulative number from one provider against a live number from another is not comparing anything.
The questions that separate providers
Once you know what the headline figure counts, three follow-ups do the real work.
How deep is it where you actually work? A global total is an average across 195 countries, and averages hide everything. If your job runs in Germany, a provider with a huge worldwide pool and thin German coverage will look excellent on the homepage and disappoint in production. This is why we report per-country figures separately rather than folding everything into one number.
How wide is the network spread? Anti-bot systems do not judge an address on its own. They judge the network it belongs to, then apply that judgment to everything else on the same network. Two providers can each hand you ten thousand distinct addresses where one draws from forty carrier networks and the other from nine hundred. The second keeps working; the first gets whole ranges flagged as soon as a few addresses burn. Spread tracks whether your scraper survives the week far more closely than raw count does.
How fast do addresses come back around? At real concurrency, not at ten requests a minute. Repeat rate under load is what actually breaks scrapers, and it varies enormously between providers advertising similar pool sizes.
Those three are properties of a network that show up in your own logs within a day of pointing real traffic at it. They are also the ones we built our benchmark around.
How we report ours
We ran a benchmark across every major residential provider we could buy access to, ourselves included, using one identical method for all of them. Same countries, same concurrency, same targets, same time window, no per-provider exceptions.
Two things about how we publish it.
Every figure carries its scope. When we report a number of distinct carrier networks and cities, we name the countries that measurement covers. A network-spread figure without a stated denominator is not a finding, it is a decoration, and the industry is full of them.
And the scores are computed from the runs rather than assigned. If a provider beats us on a metric, that is what the table says. A benchmark that always ranks its author first is an advertisement wearing a lab coat, and buyers can smell it.
Evaluating a provider in a week
If you are choosing between vendors, the fastest way past the marketing is to make them prove it on your workload.
Buy the smallest plan from your two or three finalists and run the same real job through each for a week. Not a synthetic test, the actual thing you need to do. Then compare what comes out of your own logs: what share of requests succeeded, how the failures split between timeouts and blocks and authentication errors, what the tail latency looked like rather than the median, and how much bandwidth the job burned at each provider, since per-GB pricing means an inefficient network charges you twice for the same work.
A week of your own data settles the question that no homepage can, ours included. It is also cheap. If we are one of your finalists, you can buy residential proxies on the smallest plan we sell and have a real job running against it the same afternoon.
The short version
Pool size is a real signal. Ask what it counts, ask how much of it sits in your target country, ask how many networks it spans, and then make the shortlist prove it on a week of your own traffic.
We have been running a residential network since 2012, which is long enough to have watched the headline numbers across this industry grow faster than the networks behind them. The figures worth trusting are the ones published with their method attached, which is how we run our residential proxy benchmarks.