Skip to content
Proxy fundamentals

Residential proxy vs VPN: which one do you need?

A VPN tunnels every app to a datacentre server. A residential proxy sends chosen requests out via home ISP lines. We checked 88 addresses.

On this page 7 sections
  1. Residential proxy vs VPN at a glance
  2. What an IP database sees from each
  3. The differences that bite in practice
  4. A proxy covers what you point at it, a VPN covers the device
  5. Sites can recognise a VPN from the address alone
  6. A residential exit is somebody else’s network
  7. What each operator can see
  8. UDP, calls, games and HTTP/3
  9. Billing and speed
  10. When a VPN is the right choice
  11. When a residential proxy is the right choice
  12. Check which one you are on
  13. How we tested

A VPN and a residential proxy both change the address a website sees. They get there in different ways. A VPN wraps every packet your device sends in an encrypted tunnel to one server, and that server nearly always sits in a hosting company’s rack. A residential proxy carries only the requests you point at it, and lets them out through somebody’s home broadband or mobile line. Who owns that exit address decides most of what follows.

We checked it on 21 October 2026. An IP-intelligence database labelled all 45 VPN server addresses we sampled as both VPN and hosting. It labelled none of 43 residential proxy exits as either. Seven of those 43 were already on its books as proxies, so residential does not mean unseen.

Residential proxy vs VPN at a glance

Residential proxy VPN
What it routes Only the requests you configure: a curl flag, an environment variable, a library or browser setting Every app on the device, through a virtual network interface
Who owns the exit address A consumer ISP or mobile carrier, on a real subscriber’s line A hosting or CDN company
How an IP database labelled it (our sample) 0 of 43 VPN, 0 of 43 hosting, 7 of 43 proxy 45 of 45 VPN and hosting
How the address changes Per request or per session, chosen in the username or port Fixed per server until you switch servers
Who else uses the address The household or carrier customers behind it, and other customers of the proxy pool Other subscribers on the same server
Encryption None of its own. HTTPS stays end to end inside the tunnel Everything between device and VPN server
Traffic it carries TCP, through HTTP CONNECT or SOCKS5 Any IP traffic, UDP included
Usual billing Metered per gigabyte Flat subscription
Set-up Credentials in each client, nothing installed A client app, or a WireGuard or OpenVPN profile
Pick it for Automated requests to sites that refuse datacentre addresses Your own browsing and apps on networks you do not trust

What an IP database sees from each

We sent 45 requests through one commercial rotating residential gateway: 15 each for the United States, the United Kingdom and Germany, each on a fresh session ID so it would draw a new exit. 43 came back with an address, and all 43 were different. For the VPN side we took 45 server addresses from the public lists that Mullvad, NordVPN and IVPN publish, again 15 per country, picked at random with a fixed seed. All 88 addresses then went through the same two lookups. Team Cymru’s IP-to-ASN service names the network that announces each address. The proxycheck.io v3 API assigns a network type and a set of detection flags.

Network type proxycheck.io assigned to each address, 21 October 2026 Drawn from the table below.
Network type proxycheck.io assigned to each address, 21 October 2026. Stacked bar chart of Residential, Business, Wireless, Hosting by Sample.2 rows, drawn from the table that follows this chart.Residential proxy exits (n=43)Residential proxy exits (n=43) · Residential: 2424Residential proxy exits (n=43) · Business: 1414Residential proxy exits (n=43) · Wireless: 55VPN server addresses (n=45)VPN server addresses (n=45) · Hosting: 45450204060 addressesNetwork type proxycheck.io assigned to each address, 21 October 2026. Stacked bar chart of Residential, Business, Wireless, Hosting by Sample.2 rows, drawn from the table that follows this chart.Residential proxyexits (n=43)Residential proxy exits (n=43) · Residential: 2424Residential proxy exits (n=43) · Business: 1414Residential proxy exits (n=43) · Wireless: 55VPN server addresses(n=45)VPN server addresses (n=45) · Hosting: 45450204060 addresses
Sample Residential Business Wireless Hosting
Residential proxy exits (n=43) 24 14 5 0
VPN server addresses (n=45) 0 0 0 45

Every VPN address was typed as hosting, and no residential exit was. The residential labels are not tidy, though. proxycheck called 14 exits “Business”, including lines on AT&T, Cox, BT and Telefónica Germany, all of which sell home broadband too. Treat the type label as one database’s opinion. The hosting column is the part a filter acts on.

Check Residential proxy exits (43) VPN server addresses (45)
Flagged as VPN 0 45
Flagged as hosting 0 45
Flagged as proxy 7 0
Risk score assigned 0 for 36, 100 for 7 50 for all 45
Distinct networks (ASNs) 23 8
Geolocated to the requested country 43 45
Registry allocation in the requested country 42 14

The residential exits came from 23 different access networks. Vodafone Germany supplied 6, Comcast 5, Virgin Media 5 and Deutsche Telekom 4, with T-Mobile USA and Starlink among the rest. The 45 VPN addresses came from 8 networks, every one a hosting or CDN operator: Datacamp’s CDN77 and CDNEXT networks (13 between them), 31173 Services (8), M247 (7), Tzulo (6), Clouvider (6), Hydra Communications (4) and Leaseweb (1).

The last row deserves a second look. Geolocation put every VPN address in the country its service advertised. The registry record often disagreed: British servers on address space allocated to a Swedish company, German servers on Romanian and British space. A filter that reads only geolocation sees the right country. One that compares geolocation with the allocation sees a mismatch on 31 of 45. Among the residential exits the only mismatch was a Starlink line in the UK, on SpaceX space registered in the US.

proxycheck’s documentation sets base risk scores of 33 for hosting, 50 for VPN and 100 for proxy. So when a residential exit does get caught, it scores worse than a VPN. The seven flagged exits had been seen between 21 and 443 times before our lookup.

The differences that bite in practice

A proxy covers what you point at it, a VPN covers the device

A proxy is a setting inside one program. Here is the same shell, a minute apart, without and with the flag (curl 8.7.1 on macOS 26.6.2, addresses masked):

$ curl -s https://api.ipify.org; echo
<our own address>
$ curl -s -x "http://USER-country-gb-sid-demo1:PASS@gw.example.net:5000" https://api.ipify.org; echo
81.109.x.x

Everything else on the machine still goes out directly: the next curl without -x, the system resolver, a mail client, a game launcher. Environment variables widen the net to the programs that read them, and only those, as HTTP_PROXY, HTTPS_PROXY and NO_PROXY explained shows. Name resolution is the usual leak. With an HTTP proxy the hostname travels in the CONNECT line, so the gateway resolves it. With SOCKS, curl’s socks5:// resolves locally while socks5h:// hands the name to the proxy, and the DNS leak entry explains why that matters. A browser set to use a proxy can still expose your local address through WebRTC. The curl proxy guide covers the flags in full.

A VPN sits below the applications. On WireGuard, the line that makes it a full tunnel is AllowedIPs:

[Interface]
PrivateKey = <client private key>
Address = 10.64.0.2/32
DNS = 10.64.0.1

[Peer]
PublicKey = <server public key>
Endpoint = vpn.example.net:51820
AllowedIPs = 0.0.0.0/0, ::/0

The wg-quick manual says it turns each allowed IP range into a route, and when one of them is a default route (0.0.0.0/0 or ::/0) it overrides the default gateway with a policy rule. Leave out ::/0 and IPv6 traffic bypasses the tunnel. Leave out the DNS line and a resolver on your home router keeps answering outside the tunnel, because the local network route is more specific than the default.

Sites can recognise a VPN from the address alone

VPN server lists are public by design, because the client apps need them. We downloaded all three for this test without an account. Mullvad’s list names the hosting company behind each of its 532 active WireGuard relays: Tzulo for 126, DataPacket for 122, M247 for 87, and so on, with 112 marked as owned hardware. Any IP database can import that. Ten of our 45 VPN addresses share the same proxycheck first-seen timestamp, to the second, which reads like a list loaded in one batch rather than behaviour observed over time.

Exit addresses are shared, too. In its post on exit-IP fingerprinting, Mullvad writes that “lots of users are assigned to every exit address”. Whatever the busiest user on that address did last hour is now part of its reputation. A site that refuses hosting networks refuses every VPN user without looking at anything else.

Residential exits are harder to list, because they belong to ordinary subscribers and change constantly. Harder is not impossible. Seven of our 43 were already flagged.

A residential exit is somebody else’s network

One German request failed in a way worth showing. curl refused it outright:

$ curl -sS -x "http://USER-country-de-sid-...:PASS@gw.example.net:5000" "https://api.ipify.org?format=json"
curl: (60) SSL certificate problem: self signed certificate in certificate chain

We repeated it once with verification off, to see what was at the other end. The request carried nothing private.

*  subject: CN=ipify.org
*  issuer: C=DE; ST=S-H; L=<redacted>; O=<redacted>; OU=IT; CN=CA_https
*  SSL certificate verify result: self signed certificate in certificate chain (19), continuing anyway.
< HTTP/1.1 403 Forbidden
< Via: 1.1 <redacted>.local (squid)

That exit was a line inside an organisation whose web filter, a Squid proxy, re-signs HTTPS certificates with its own authority and then decides what its users may see. On a clean exit the same certificate came back issued by Google Trust Services (CN=WE1). We have removed the organisation’s name. It was a real network with its own rules, and the only thing that kept our request out of someone else’s TLS interception was certificate verification.

Never switch certificate checks off to make a residential proxy “work”. No -k in curl, no verify=False in Python, no rejectUnauthorized: false in Node. A certificate error through a residential exit usually means exactly what it says. Retry on a new session instead.

Where the exits come from is another thing you cannot see from the client side. A 2019 IEEE Symposium on Security and Privacy paper by Mi and colleagues found 6 million residential proxy addresses, and reported that many ran on likely compromised hosts, IoT devices among them, despite providers’ claims that the hosts joined willingly. A VPN server is a machine the VPN company rents or owns. You still have to trust that company, but you know who it is.

What each operator can see

Neither tool hides you from the people running it. This is the start of a clean proxied request in verbose mode:

* Proxy auth using Basic with user 'USER-country-gb-sid-demo24918'
* Establish HTTP proxy tunnel to api.ipify.org:443
> CONNECT api.ipify.org:443 HTTP/1.1
> Host: api.ipify.org:443
> Proxy-Authorization: Basic <redacted>
> User-Agent: curl/8.7.1
> Proxy-Connection: Keep-Alive
>
< HTTP/1.1 200 Connection established
<
* CONNECT phase completed
* SSL connection using TLSv1.3 / AEAD-CHACHA20-POLY1305-SHA256 / [blank] / UNDEF
* ALPN: server accepted h2
*  issuer: C=US; O=Google Trust Services; CN=WE1
*  SSL certificate verify ok.
* using HTTP/2

The gateway learns the hostname and port from the CONNECT line, and your credentials from the Proxy-Authorization header. Basic credentials are only base64-encoded (RFC 7617), and with an http:// proxy URL the hop to the gateway is not encrypted, so anyone on that path can read them. Inside the tunnel the TLS session runs end to end with the site. The gateway and the exit see bytes, not pages.

A VPN provider sees the same kind of metadata for every app at once: destination addresses, the server name in each TLS handshake, and your DNS queries if you use its resolver. Plain HTTP is readable and changeable by either operator. What a VPN adds is the encrypted hop from your device to its server (ChaCha20 with Poly1305 on WireGuard), which is what protects you on hotel or café Wi-Fi.

UDP, calls, games and HTTP/3

WireGuard encapsulates IP packets over UDP, so anything that speaks IP goes through it: DNS, voice calls, games, QUIC. An HTTP proxy opens a TCP tunnel with CONNECT (RFC 9110, section 9.3.6), and TCP is all it carries. HTTP/3 runs on QUIC over UDP, so a client behind a CONNECT proxy falls back to HTTP/2 or 1.1. Our run above negotiated h2. UDP over a proxy does exist on paper. SOCKS5 has UDP ASSOCIATE (RFC 1928) and HTTP has CONNECT-UDP (RFC 9298). Whether a particular residential gateway supports either is a question for its documentation.

Billing and speed

VPN subscriptions are usually flat, so a day of video costs the same as a day of email. Residential plans are usually metered per gigabyte, and every byte the exit relays is billable. The classic expensive mistake is a headless browser fetching images and fonts it never needed.

Each residential request also crosses a stranger’s last mile. Timings for our 43 successful requests to a JSON endpoint that returns under 30 bytes:

Median time per request through the residential gateway, 21 October 2026 Drawn from the table below.
Median time per request through the residential gateway, 21 October 2026. Bar chart of Median total time (s) by Exit country.3 rows, drawn from the table that follows this chart.00.250.50.7511.25 sUnited States: 1.1611.161United StatesUnited Kingdom: 0.5730.573United KingdomGermany: 0.6020.602GermanyMedian time per request through the residential gateway, 21 October 2026. Bar chart of Median total time (s) by Exit country.3 rows, drawn from the table that follows this chart.United StatesUnited States: 1.1611.161United KingdomUnited Kingdom: 0.5730.573GermanyGermany: 0.6020.60200.511.5 s
Exit country Successful requests Median total time (s)
United States 14 1.161
United Kingdom 15 0.573
Germany 14 0.602

Across all 43 the median was 0.73 s, the 90th percentile 1.44 s and the slowest 4.33 s. TCP connect to the gateway took a median 0.036 s from our test machine in Europe, so nearly all the time went on the residential hop and the TLS handshake through it. One US session failed after 13.5 s. We did not time a VPN on the same machine, so there is no VPN figure to set beside these.

When a VPN is the right choice

Use a VPN when the traffic is yours.

  • Your laptop or phone is on Wi-Fi you do not control, and you want every app covered, not only the browser.
  • You want your ISP or the local network to see one encrypted stream instead of your destinations.
  • You use apps that need UDP: calls, games, anything on QUIC.
  • You move a lot of data and want a flat price.
  • You log in to your own accounts. Routing your bank through an unknown residential line adds a party you cannot vet.

Expect sites with strict filters to treat you as a VPN user. Every server address in our sample was labelled as one.

When a residential proxy is the right choice

Use a residential proxy when the traffic is a program’s and the target refuses datacentre addresses.

  • Scrapers, price checks and ad verification that need many addresses, or a new one per request.
  • Checking what a page shows in a given country, city or carrier, from an address that looks local.
  • Multi-step flows that need one address for several minutes. Use a sticky session, as rotating vs sticky sessions explains.

Try a datacenter proxy first. If the target does not check network origin it is faster and cheaper, and the residential vs datacenter guide shows how to find out. If a clean residential exit still gets a 403, the address was probably not the problem. 403 vs 407 vs 429 and our TLS fingerprint research cover what else gives a client away.

Check which one you are on

The quickest test is to ask who announces your address. Team Cymru’s whois service answers in one line. Here it is for a NordVPN server address from the public list:

$ whois -h whois.cymru.com " -v 89.35.28.131"
AS      | IP               | BGP Prefix          | CC | Registry | Allocated  | AS Name
25369   | 89.35.28.131     | 89.35.28.0/22       | GB | ripencc  | 2005-11-29 | BANDWIDTH-AS - Hydra Communications Ltd, GB

And this is the line our bulk lookup returned for one residential exit, address masked:

7922    | 73.x.x.x         | 73.0.0.0/8          | US | arin     | 2005-04-19 | COMCAST-7922 - Comcast Cable Communications, LLC, US

To check your own connection, with the VPN up or down:

whois -h whois.cymru.com " -v $(curl -s https://api.ipify.org)"

A hosting company in the last column means a filter will see a datacentre, whatever the product is called. If something is sold to you as a “residential VPN”, this lookup tells you whose network its exit is really on. The ASN entry explains what the columns mean.

How we tested

All runs took place on 21 October 2026 between 19:27 and 19:39 UTC, from a macOS 26.6.2 machine with curl 8.7.1, DiG 9.10.6 and Python 3.15.0.

  • Residential sample. 45 requests to https://api.ipify.org?format=json through one commercial rotating residential gateway, 15 per country (US, GB, DE), five at a time, each on a fresh session ID. 43 returned an address and all 43 were distinct. We do not name the provider, because this page is not a review of it. We do not publish the exit addresses either: they belong to subscribers.
  • VPN sample. Server lists fetched the same day from Mullvad’s relay API, NordVPN’s server API and IVPN’s server list, none of which needed an account. Up to five IVPN servers per country, the rest split between Mullvad and NordVPN, drawn with Python’s random seeded with 20261021: 17 Mullvad, 16 NordVPN and 12 IVPN addresses.
  • Lookups. One bulk query to Team Cymru’s whois service and one keyless request to proxycheck.io v3, each covering all 88 addresses. We also pulled each network’s self-declared type from PeeringDB. It did not separate the two groups at all: AT&T, Deutsche Telekom and M247 all describe themselves as “NSP”.

The per-request script, with credentials read from the environment:

#!/bin/bash
# usage: res_one.sh <country> <session-id>  -> country,session,exit_ip,code,connect,appconnect,total
cc="$1"; sid="$2"
out=$(curl -sS -m 30 -x "http://${PROXY_USER}-country-${cc}-sid-${sid}:${PROXY_PASS}@gw.example.net:5000" \
  -w '\t%{http_code}\t%{time_connect}\t%{time_appconnect}\t%{time_total}' \
  "https://api.ipify.org?format=json" 2>&1 | tr '\n' ' ')
ip=$(printf '%s' "$out" | sed -n 's/.*"ip":"\([^"]*\)".*/\1/p')
rest=$(printf '%s' "$out" | awk -F'\t' '{print $2","$3","$4","$5}')
printf '%s,%s,%s,%s\n' "$cc" "$sid" "${ip:--}" "$rest"

Run it from a separate shell, not from inside the script, with 15 sessions per country and five in parallel:

for cc in us gb de; do for i in $(seq 1 15); do echo "$cc r${cc}${i}x$RANDOM$RANDOM"; done; done \
  | xargs -P 5 -n 2 ./res_one.sh > res_exits.csv

The bulk ASN lookup. Piping the same query into nc returned nothing for us, because it closed before Team Cymru answered; a socket that waits for the server to hang up works:

import socket, sys
ips = open(sys.argv[1]).read().split()
s = socket.create_connection(("whois.cymru.com", 43), timeout=30)
s.sendall(("begin\nverbose\n" + "\n".join(ips) + "\nend\n").encode())
out = b""
while chunk := s.recv(65536):
    out += chunk
print(out.decode(), end="")

The classification call. A POST to /v3/ with no query string was redirected to a lookup of our own address and ignored the body. Adding a query string (we used ?ver=1) made it read the list:

curl -sS -d "ips=$(paste -sd, ips.txt)" "https://proxycheck.io/v3/?ver=1"

The limits are real. This is one residential provider, one classifier and one evening’s snapshot. The VPN addresses are the ones a client connects to. Mullvad says each of its servers holds several exit addresses, so an exit can differ from the listed address while sitting on the same machine. A paid IP database or a different residential pool would give different counts. The direction of the result, hosting labels on every VPN address and on no residential exit, is the part we expect to hold.

Frequently asked questions

Is a residential proxy a VPN?
No. A VPN creates an encrypted tunnel for all of a device's traffic to a server, usually in a datacentre. A residential proxy relays only the requests an application sends to it, out through a home or mobile connection, and adds no encryption of its own.
Can websites tell that I am using a VPN?
Often from the address alone. In our 21 October 2026 sample, proxycheck.io labelled all 45 published VPN server addresses as VPN and hosting. VPN server lists are public, and many subscribers share each exit address, so its reputation is everyone's.
Are residential proxies undetectable?
No. The same database labelled none of our 43 residential exits as VPN or hosting, but 7 were already flagged as proxies and scored 100, the top base risk score in its scheme. Your client's TLS fingerprint and behaviour can give you away whatever the address.
Is a residential proxy safer than a VPN?
For your own browsing, no. The exit is a stranger's connection you cannot vet, and one of our exits intercepted TLS with its own certificate authority. A VPN encrypts everything to a server run by a company you chose. With either, keep certificate verification switched on.
Can I use a residential proxy for streaming, calls or games?
Not well. An HTTP proxy carries TCP through CONNECT, so UDP traffic such as voice, games and HTTP/3 does not pass, and per-gigabyte billing makes video expensive. A VPN carries any IP traffic, UDP included, and is usually billed at a flat rate.
How do I check whether an address is residential or a datacentre?
Look up the network that announces it. Running whois -h whois.cymru.com " -v ADDRESS" prints the AS name: a consumer ISP or mobile carrier means a residential exit, a hosting or CDN company means a datacentre. That works for anything sold as a residential VPN too.

Sources

  1. Mullvad relay list API (fetched 21 October 2026)
  2. NordVPN server list API (fetched 21 October 2026)
  3. IVPN server list (fetched 21 October 2026)
  4. Team Cymru: IP to ASN mapping
  5. proxycheck.io API documentation (v3 fields and risk score base values)
  6. PeeringDB API documentation
  7. Mullvad: Exit IP fingerprinting between VPN servers
  8. WireGuard: Protocol and cryptography
  9. wg-quick(8) manual page
  10. RFC 9110: HTTP Semantics, section 9.3.6 CONNECT
  11. RFC 7617: The Basic HTTP Authentication Scheme
  12. RFC 1928: SOCKS Protocol Version 5
  13. RFC 9298: Proxying UDP in HTTP
  14. curl manual: --proxy and SOCKS schemes
  15. Mi et al.: Resident Evil, Understanding Residential IP Proxy as a Dark Service (IEEE S&P 2019)

Read this page as Markdown · Quote it freely under CC BY 4.0 with a link back.