More unique IPs than any network on the planet, in the world's largest residential pool. That's Shifter.More unique IPs than any network, in the world's largest pool. That's Shifter.

See the numbers

Fastest Residential Proxies: How to Judge Speed Claims

What "fastest proxies" and proxy ping figures actually measure, the questions that expose a flattering speed claim, and how Atlas turns latency into a speed score you can check.

Chris Collins Engineering · Updated November 11, 2026 · 7 min read
net 1p99net 2net 3net 4fastslow
Latency per network. The best median has the longest amber tail.
Performance
On this page
  1. What a proxy ping measures
  2. Five questions that expose a speed claim
  3. A fast block is still a failure
  4. How Atlas scores speed
  5. Speed varies by target and by country
  6. When speed matters, and when it does not
  7. Checking a speed claim yourself
  8. Frequently asked questions

Every proxy provider has a speed claim. Some quote a response time, some a “ping”, some just say “fastest”. These figures are rarely wrong, exactly. They are measured in the most favourable way available, and then compared with figures that were measured differently.

If you are choosing residential proxies partly on speed, the useful skill is not finding the smallest number. It is knowing what each number measured, so you can tell which ones describe the requests your code will actually send.

What a proxy ping measures

“Proxy ping” usually means one of two things: an ICMP ping to the provider’s gateway, or the time to open a connection to it. Both measure the distance between you and the provider’s entry server. Neither touches the part of the trip that makes residential proxies slow.

A residential request travels from your client to the gateway, from the gateway to a consumer device on a home or mobile connection, from that device to the target, and back the same way. The gateway is usually in a well-connected data centre close to you. The exit is a device on someone’s home broadband or Wi-Fi, which may be on another continent. A low ping to the gateway is good to have, but it is the smallest share of what you wait for.

So a ping is a fine diagnostic for “is my connection to the provider healthy?” and a poor answer to “which provider is fastest?”.

Five questions that expose a speed claim

When you see a speed figure, from a provider, a review or a benchmark, ask:

  1. Measured to what point? Connect time, first byte or last byte. Last byte is what your code waits for.
  2. Which percentile? An average or a median hides the slow tail. The guide to reading a latency chart explains why the p99 often matters more.
  3. Under what load? Ten requests on an idle account look nothing like a thousand concurrent ones.
  4. From where, to where? The vantage point, the exit country and the target all move the result. A figure for one country says little about another.
  5. What happened to failures? If blocks and timeouts were dropped before averaging, the figure describes only the requests that went well.

A claim that cannot answer these is a marketing number. That does not make it false; it makes it impossible to compare.

A fast block is still a failure

The most common way to look fast is to fail quickly. A target that refuses a request with a 403 or serves a CAPTCHA does less work than one that renders the full page, so the refusal often comes back sooner. A pool full of flagged addresses can post an impressive response time for exactly that reason.

That is why speed only means something next to success. Atlas scores each target as 80% success and 20% speed under load, each against the leader, so a network cannot climb by being fast at being refused. The same principle applies to your own testing: never read a timing column without the success column beside it. For what slow and failed requests do to your bill, see cost per successful request.

How Atlas scores speed

Atlas publishes one speed figure in the score and shows the rest beside it.

What is timed. The time from request start to last byte, through the proxy, on a plain-HTTP target. HTTPS targets add a CONNECT and a TLS handshake, roughly doubling the round trips, so they report success and blocks but not latency. Mixing them in would reward networks for which targets they happened to be tested on.

Under load. Each collector region picks its fastest country as a reference. In every time slot that country gets a light pulse of requests and, a minute later, a wave at full census concurrency. The ratio between the two is the region’s slow-down factor, applied to the light-load latency of every country tested from that region. Light-load latency is shown next to it, so you can see how much each network slows when it is busy.

Scored as a ratio. The speed sub-score is the network’s under-load p50 set against the fastest network in Atlas: 100 × leader ÷ yours. The leader scores 100; a network at twice the leader’s latency scores 50. Speed carries 14% of the default Atlas Score. The exact scoring rules show every step.

If speed matters more to your workload than the default weights assume, the leaderboard’s Latency first preset re-weights the ranking so speed counts for more.

Speed varies by target and by country

There is no single “fastest residential proxy”. There is a network that is fastest from a given region, to a given country, on a given kind of target, at a given time of day.

Country. Exits in countries with fast home broadband and short routes to the target answer sooner. Asking for a narrow location, such as a city, shrinks the set of exits to choose from, and the gateway may hand you a slower one because it is the only one available.

Target. A heavy page with large assets takes longer than a small JSON endpoint on any network. Defended targets may add a challenge round trip even on a request that eventually succeeds. The per-target results show how much the same network’s outcome changes from one site to the next.

Time of day. Residential exits share bandwidth with the people who own the devices. Evening peaks in the exit’s country can slow requests that ran smoothly at breakfast. A figure from a single hour is a snapshot.

When speed matters, and when it does not

Before paying more for speed, check whether your job is limited by it.

Bulk collection. If you are fetching a large list of pages with no user waiting, throughput is set by concurrency more than by per-request latency. Running more requests in parallel usually finishes sooner than switching to a network with a slightly lower median. Success rate matters more here, because every failure is a retry.

Interactive or real-time work. Price checks at the moment of a decision, ad verification, or anything a person is waiting on is limited by the tail. Here the p90 and p99 matter as much as the median.

Multi-step sessions. Logins, carts and paginated flows run in sequence, so each slow step adds to the total. Consistency matters more than raw speed.

Much of the latency people blame on providers is in their own client: no connection reuse, over-tight geo filters, long timeouts. Reducing latency when using residential proxies covers the fixes worth trying before you switch.

Checking a speed claim yourself

  1. Pick two or three targets you actually use, including at least one plain-HTTP page so you can time the network without the TLS round trips.
  2. Send the same requests through each network in the same hour, from the same machine, at the concurrency you run in production.
  3. Time each request to the last byte, and record whether it passed your own check for a real page.
  4. Report the p50, p90 and p99 of successful requests, with the success rate next to them.
  5. Repeat at a busy hour in the exit country. If the ranking flips, you have learned which network holds up under load.

Then compare your figures with Atlas, which runs the same discipline continuously and tests Shifter by the same rules as every other network.

Frequently asked questions

Are residential proxies with lower latency always faster?

Not necessarily. A refusal or a CAPTCHA often comes back sooner than a real page, so a pool of flagged addresses can post a low figure. Always read a speed figure next to the success rate on the same target.

How can I test the speed of a residential proxy?

Send the same requests to the same target, from the same location and in the same window, at the concurrency you will really run. Time each request to the last byte, record whether it returned the real page, and compare percentiles such as p50 and p99 rather than one average.

How can I tell whether a residential proxy speed claim is reliable?

Ask what was timed (connect time, first byte or last byte), which percentile is quoted, under what load, from where to which target, and what happened to failed requests. A figure that cannot answer these is not necessarily false, but it cannot be compared with anything else.

Get started

Time Shifter on your own targets

Start a free trial of Shifter's residential network and measure end-to-end time on the pages you actually fetch, at the concurrency you actually run.

3-day money-back guarantee. Cancel anytime.