Multi-Vantage IP Check

Compare the address our anycast edge sees with the address our IPv4-only and IPv6-only hostnames see, and read what any difference means.

What each row is, exactly. The first three rows are hostnames your browser contacted, so each one reports the address it actually saw. The probe rows are different: your browser never contacts a probe, and we never forward your address to one, so a probe can only report its own egress address. Nothing on this page claims a probe saw you.

asking the single-stack hostnames what they see

Vantage pointAddress it sawNetworkLocationWhat this row can and cannot show
Our edgethis page216.73.216.7AS16509 Anthropic, PBCColumbus, USThe address your browser used to reach us. The location is the edge that answered (CMH), and our edge exists in many cities, so this location can move without your address changing.
IPv4-only hostnamev4 originasking…A separate hostname that answers over IPv4 only. A different address here than at the edge means your traffic leaves by more than one path.
IPv6-only hostnamev6 originasking…Blank means this connection has no working IPv6, which is a fact about your path and not an error.

Loading the probe roster.

Compare in this order: address, then network, then location. The address is an observation. The network is a lookup in routing data and is nearly always right. The location is an estimate and is the least reliable field on the page, so a location difference on its own means nothing.

What each row is, exactly. The first three rows are hostnames your browser contacted, so each one reports the address it actually saw. The probe rows are different: your browser never contacts a probe, and we never forward your address to one, so a probe can only report its own egress address. Nothing on this page claims a probe saw you.

Why one vantage point is not enough

Every “what is my IP” page, this site included, answers the question “what address did you use to reach this server”. That is one measurement. It is correct and it is incomplete, because the answer can depend on the destination.

Three cases make that difference visible, and all three confuse people regularly.

  • CGNAT pools. One subscriber, several public addresses, chosen per connection.
  • Split-tunnel VPNs. Different destinations, different egress.
  • Anycast edges. One address, many physical locations, so the reported city changes without the address changing.

What the probe rows are, and are not

Our regional probes appear on this page, and they are not vantage points on you. Your browser never contacts one, and the Worker never forwards your address to one in any header or field. A probe can only report the address the outside world sees for the probe itself. The row is here so you can tell our machines apart from yours; nothing on this page claims a probe observed your connection.

Reading the comparison table

Compare in this order: address first, then ASN, then location. The address is an observation. The ASN is a lookup in routing data and is nearly always right. The location is an estimate and is the least reliable field on the page, so a location difference alone means nothing.

Limits

Three vantage points is enough to detect that paths differ. It is not enough to map which destinations take which path. If you need that, the answer is a routing table on your own machine, not a web page.

About this tool

Your browser fetches the same JSON payload from our anycast edge and from two single-stack hostnames, one IPv4-only and one IPv6-only, and we show the address, network, and approximate location each one observed. Usually all three agree. When they disagree, the difference says something specific about your path, such as a CGNAT pool or a split tunnel, and the page names the likely cause.

How to read the result

All three agree
All three see one egress address. This is the normal case for a residential connection with no VPN, and also for a correctly configured full-tunnel VPN.
Edge differs from both hostnames
The address is the same but the observed location differs, because our edge sits on an anycast network with points of presence in many cities and the single-stack hostnames do not. You reached a different edge, not a different address.
Different addresses from the same ISP prefix
Your carrier is load balancing across a NAT pool, which is standard CGNAT behaviour on mobile and on some home connections. Each new connection can take a different public address from the same block, which breaks any service that assumes one session equals one address.
Different addresses from different ASNs
Traffic is splitting across paths, which on a VPN means split tunnelling, where some destinations go through the tunnel and others do not. It is the same class of problem as a WebRTC or DNS leak, seen at the transport layer.
One hostname unreachable
One of the three did not answer, so we report it as unreachable rather than folding it into the comparison, and the cause can be a network filter, an outage on our side, or an extension blocking a cross-origin request. Two agreeing answers and one failure is still a usable result, and one answer alone is not.
IPv4 and IPv6 seen separately
One hostname answers over IPv4 only and one over IPv6 only, so each row names the family that carried it. An IPv6 row on a different network from the IPv4 row is the classic dual-stack VPN leak, where the client tunnels one family and ignores the other.

Questions people ask

Why do two websites show me different IP addresses?
Three usual causes. Carrier-grade NAT assigns your connections different public addresses from a pool. A split-tunnel VPN routes some destinations through the tunnel and others directly. Or one site reads a proxy header incorrectly. This page distinguishes the first two by comparing the ASN of each observed address.
Is a different location on each row a problem?
Not by itself. Geolocation databases differ, and our edge is anycast, so the location reflects which point of presence answered. Compare the addresses and the networks. Those are facts. Locations are estimates from different sources and will differ routinely.
How is this different from the VPN check?
The VPN check tests the channels that leak around a tunnel: WebRTC, DNS, and IPv6. This one tests whether ordinary HTTPS traffic to different destinations leaves by different paths. A split tunnel that keeps DNS and WebRTC inside the tunnel passes the VPN check and fails here.
My mobile connection shows a different address on every reload. Why?
Carrier NAT pools. Mobile networks put many subscribers behind a set of public addresses and assign them per connection, sometimes per flow. You cannot change it, and services that pin a session to an address will log you out unexpectedly because of it.
Do the regional probes see my address?
No, and they cannot. Your browser never contacts a probe, and we never forward your address to one in any header or field. A probe row on this page reports the probe's own egress address, so you can tell our vantage points apart from yours. Every row that reports an address you used is one of our own hostnames that your browser contacted directly.

Related

Last reviewed 2026-09-05.