IPv6 is not a new idea, but it is still unevenly deployed. Two networks can both claim to support it while behaving very differently: one may have a working address on the router, the other may have a route that is advertised but blackholed. The only reliable way to know is to test the exact path you care about.
This guide explains the protocol in plain language, then walks through testing and enabling it on routers, servers and websites.
What IPv6 is
IPv6 is the current version of the Internet Protocol. It replaces the 32-bit addresses of IPv4 with 128-bit addresses, written as eight groups of hexadecimal digits and shortened with :: where a run of zeros appears. A typical address looks like 2a02:8108:50bf:c694::1.
The address space is large enough that every device can have a globally unique address without being multiplexed behind a shared IPv4 address. IPv6 also changes how networks configure themselves: routers advertise prefixes, hosts form addresses with stateless address autoconfiguration, and Neighbor Discovery replaces ARP. Broadcast is gone; multicast and anycast take its place.
Not every IPv6 address is meant for the public internet. Global unicast addresses come from 2000::/3. Link-local addresses from fe80::/10 stay on one link. Unique local addresses from fc00::/7 are private, and multicast uses ff00::/8. A website needs a global unicast AAAA record; a link-local address on a server is not enough.
IPv6 address examples
An IPv6 address has 128 bits, written as eight groups of 16 bits. Each group has four hexadecimal digits, and :: compresses one run of zero groups. Two examples:
2001:db8::1comes from the documentation prefix2001:db8::/32, reserved for examples and not routable.2a00:1450:4003:80b::200eis a real-world global unicast address of the kind a public AAAA record returns.
:: can appear only once, because two compressed runs would be ambiguous. The IPv6 details that matter on a LAN are the prefix and its length: a /64 is common because stateless address autoconfiguration expects a 64-bit interface identifier, and it leaves 64 bits for subnets.
Why IPv6 matters
IPv4 has roughly 4.3 billion addresses, and the public pool was exhausted years ago. Networks kept growing by sharing addresses: carrier-grade NAT at mobile operators, NAT at home and in offices, and port forwarding when something needs to be reachable from outside. That works for outbound browsing and becomes fragile for inbound services, peer-to-peer traffic, gaming, voice and anything that benefits from a stable address.
IPv6 removes that pressure. Mobile networks in many countries already run IPv6 for a large share of traffic. Cloud providers offer native IPv6 subnets. A website that only publishes A records is still reachable through translation or a proxy, but it is not directly reachable from an IPv6-only client. The user may see a slower path, a broken feature, or no connection at all.
The operational argument is just as important. Fewer translation layers mean fewer state tables, fewer mysterious timeouts and a simpler path for monitoring. Dual-stack deployment lets you add IPv6 without breaking IPv4, then measure what actually changes.
Dual stack, CGNAT and NAT64
Dual stack means a device or network runs IPv4 and IPv6 at the same time. A hostname can publish both an A and an AAAA record, and a browser can try both. Modern browsers use Happy Eyeballs: they start connections over both families and keep the one that answers first. That is why a broken IPv6 path sometimes looks like a slow site rather than a failed one.
CGNAT is carrier-grade NAT. Many customers share one public IPv4 address, and the carrier tracks which internal connection owns which port. Outbound browsing works, but inbound connections generally do not. IPv6 avoids this by giving the customer a routed prefix instead of a shared address.
NAT64 and DNS64 let an IPv6-only network reach IPv4-only destinations. DNS64 synthesizes AAAA answers from A records, and NAT64 translates the packets. It is a useful transition tool, but it is still translation: it can break protocols that embed addresses, and it does nothing for inbound reachability to an IPv4-only service. A direct AAAA record is simpler and more durable.
How to check for IPv6
Start with the connection. The home page check asks the server which address family your browser used and shows your public IP address without logging it. If it reports IPv6, your current path can reach an IPv6-only service. If it reports IPv4, your provider may still offer IPv6 on another network or after a router change.
On Windows, ipconfig /all lists the addresses assigned to each adapter. On macOS, ifconfig does the same. On Linux, ip -6 addr show scope global lists the global addresses. Look for a global address that begins with 2 or 3; a link-local address beginning with fe80 is normal but is not routable on the internet.
On the router, check the WAN status for an IPv6 address and, when the ISP delegates one, a prefix for the LAN. Router advertisements must be allowed on the LAN so clients can form addresses. Then repeat the home page check from the client that matters. Do not test only the router: it can have a working IPv6 uplink while prefix delegation is broken, or the opposite.
IPv6 DNS and lookup: find a website's AAAA records
An IPv6 lookup is a DNS question: does this hostname publish an AAAA record, and which address does it contain? The AAAA record is the IPv6 equivalent of the A record, and a dual-stack site publishes both. A missing AAAA record is a DNS fact, not proof that the site is broken; it means IPv6-only clients need translation or a proxy to reach it.
IPv6 DNS uses the same resolver chain as IPv4; only the record type changes. An AAAA record maps a hostname to one or more 128-bit addresses, and a name can publish both record types. DNS can run over either family, so an IPv4 resolver can still return a valid AAAA answer.
From a terminal, query the record directly:
# Ask for the IPv6 address only
dig AAAA example.com +short
# Same query with nslookup
nslookup -type=AAAA example.com
# Compare with the IPv4 address
dig A example.com +short
The answer may start with a CNAME, especially when the site uses a CDN. Follow the chain until you reach the AAAA records. Multiple AAAA records are normal: they usually represent several edge or origin addresses.
The website checker performs the same lookup in the browser, then goes one step further and dials the address it found. That separates a DNS problem from a routing, firewall or listener problem.
Testing IPv6 connectivity
DNS answers are not reachability. A name can have a valid AAAA record while the address refuses connections, and a working IPv6 path can exist for a name that has no AAAA record. Force the connection and compare both families:
# Force IPv6 and check the HTTP status
curl -6 -sS -o /dev/null -w '%{http_code}\n' https://example.com
# Compare with IPv4
curl -4 -I https://example.com
# Send ICMPv6 echo requests (may be filtered)
ping6 example.com
If curl -6 succeeds on one network and fails on another, compare the route, the firewall and the MTU. If dig AAAA returns an address but curl -6 hangs or fails, the problem is reachability rather than DNS.
ping6 is a useful signal but not a verdict. Many networks filter ICMPv6 echo requests while allowing TCP and TLS, so a failed ping does not prove that the website is down, and a successful ping does not prove that HTTPS works. This site's checker uses TCP, TLS and one HTTP request, not ICMP.
Path MTU is another classic failure. IPv6 does not fragment in routers; it relies on ICMPv6 Packet Too Big messages. If those messages are filtered, large packets disappear while small requests still work. Test the public hostname rather than an IP address so DNS is included, and test from the client or region that reports the problem.
How to enable IPv6
On a home or office router: enable IPv6 in the WAN settings, request a delegated prefix if your ISP provides one, and allow router advertisements on the LAN. Check that the firewall permits the traffic you intend to expose and blocks the rest. Avoid disabling IPv6 on clients as a workaround for a broken router; fix the router instead.
With your ISP: confirm that IPv6 is enabled on the line and that the provided router supports it. Some ISPs enable it by default, some require a setting, and some still do not offer it. A tunnel broker can provide connectivity over IPv4, but it adds latency and another dependency.
On a VPS or dedicated server: request a native IPv6 address or prefix, configure the address and default gateway, and confirm the provider is not filtering it. Then make the service listen on IPv6. A process bound only to 0.0.0.0 is not listening on IPv6. Bind to :: or to both families explicitly, and check the firewall for ICMPv6, which is required for Neighbor Discovery and path MTU discovery.
On a web server: add an IPv6 listener. In nginx, listen [::]:443 ssl; alongside the IPv4 listener is enough when the address is configured. Apache uses Listen [::]:443. Caddy and many modern proxies handle both families automatically. Reload the service, then confirm with ss -6 -ltn that the port is bound.
In DNS: publish an AAAA record next to the A record. Keep both during the transition. If you use a CDN, enable IPv6 at the edge first: visitors then connect to the edge over IPv6 while the origin can remain IPv4-only. For direct origin reachability, enable IPv6 on the origin as well.
IPv6 on routers, firewalls and tunnels
An IPv6 router or CPE needs IPv6 enabled on the WAN, a prefix delegated by the ISP, and router advertisements on the LAN. Ask for a routed prefix, not a single address: a single address configures only the router. The LAN-side IPv6 configuration should hand out a /64 and a working DNS resolver. Avoid NAT66; it adds translation state without the end-to-end routing IPv6 is meant to provide.
An IPv6 firewall should not block the protocol by default. Allow ICMPv6 Neighbor Discovery (types 133-137) on each link and ICMPv6 Packet Too Big (type 2) on every path, or address resolution and path MTU discovery break. Permit inbound TCP 443 and 80 only on hosts that serve web traffic, and keep a default deny for the rest.
An IPv6 tunnel carries IPv6 over an IPv4 path, usually through a broker. It is useful for testing, labs, or legacy networks without native IPv6. Native dual stack is better for production: fewer hops, no broker dependency and no encapsulation overhead. If you use one, lower the MTU (1280-1480 is typical) and confirm ICMPv6 Packet Too Big reaches both ends; otherwise large responses stall.
Why a website may fail over IPv6
The most common failure is a missing AAAA record. The second most common is an AAAA record that points at an address where nothing is listening. A firewall may allow port 443 over IPv4 and silently drop it over IPv6. A load balancer may have an IPv6 front end but an IPv4-only backend. A TLS certificate may be valid for the name but the IPv6 virtual host may serve a different certificate.
Path MTU is another classic. IPv6 does not fragment in routers; it relies on ICMPv6 Packet Too Big messages. If ICMPv6 is filtered, large packets disappear and small requests may work while page loads stall. Neighbor Discovery also depends on ICMPv6, so an overzealous firewall can break the local link entirely.
Finally, caching hides fixes. DNS resolvers, CDNs and this checker all keep answers for a while. If you just changed a record or opened a firewall, wait for the TTL and test again from more than one network.
IPv4 vs IPv6: what's the difference?
The main difference between IPv4 and IPv6 is address size: 32 bits versus 128 bits. That drives the comparison below.
| Property | IPv4 | IPv6 |
|---|---|---|
| Address size | 32 bits | 128 bits |
| Notation | Dotted decimal | Hexadecimal groups with :: |
| Public addresses | Exhausted, heavily shared | Abundant, routed prefixes |
| Configuration | DHCP common | Router advertisements and DHCPv6 |
| Broadcast | Yes | No, multicast instead |
| Header checksum | Yes | No, left to upper layers |
| Minimum MTU | 68 bytes | 1280 bytes |
| NAT | Common | Not required for address shortage |
Methodology and limitations
This site performs a narrow, reproducible check. It validates the hostname, resolves AAAA and A, rejects non-global IPv6 addresses, dials only the resolved address on port 443 (then 80), completes TLS when available, sends one HTTP request to the root path. It never follows redirects. The total deadline is six seconds.
A result describes one vantage point at one moment. It does not test every subdomain, asset, API or geographic region, and it cannot see a user's local network. Results are cached for five minutes by hostname. Visitor IP addresses are used in memory for rate limiting and are not logged or stored.
Frequently asked questions
How do I check for IPv6?
Use the home page panel to see your public IP address and whether this page loaded over IPv4 or IPv6. On Windows, run ipconfig /all; on macOS, use ifconfig; on Linux, use ip -6 addr. Then confirm that your router has an IPv6 WAN address and a delegated prefix for the LAN.
How do I look up an IPv6 address?
Query DNS for the AAAA record with dig AAAA example.com +short or nslookup -type=AAAA example.com. Then test the address itself with curl -6 -sS -o /dev/null -w '%{http_code}\n' https://example.com or the website checker, which uses TCP, TLS and one HTTP request.
What is the difference between an IPv6 test and an IPv6 lookup?
A test checks whether a connection or service works over IPv6: a reachability question. A lookup finds the IPv6 address published in DNS: a resolution question. A hostname can have a valid AAAA record and still fail the connection test.
How do I test if a website supports IPv6?
Enter the hostname in the website checker. It reads the AAAA record, then opens a real IPv6 connection on port 443 (falling back to 80), so DNS and reachability are reported separately. A hostname without AAAA is IPv4-only in DNS.
Is IPv6 faster than IPv4?
Not inherently. IPv6 removes some NAT overhead and can take a more direct path, but routing, peering, latency and server load matter more. Happy Eyeballs lets browsers race both families and use the faster connection.
What is the difference between IPv4 and IPv6?
IPv4 uses 32-bit addresses (about 4.3 billion) and networks stretch them with NAT. IPv6 uses 128-bit addresses, so every device can have a globally unique address. IPv6 also drops broadcast, uses router advertisements for configuration, and leaves fragmentation to hosts instead of routers.
Do I need to disable IPv4 to use IPv6?
No. Run both protocols in dual-stack mode. Disabling IPv4 would cut off users and services that still depend on it. The goal is to add IPv6, not to remove IPv4 yet.
Why do some sites only work over IPv4?
The hostname may lack an AAAA record, the IPv6 listener may be missing, a firewall may block port 443, or the route may be broken. A website can have IPv6 at the CDN edge or at the origin, and those are separate layers.
How do I test IPv6 without this website?
Use dig AAAA example.com to check DNS, then curl -6 -sS -o /dev/null -w '%{http_code}\n' https://example.com to force an IPv6 connection. Compare it with curl -4. On Linux, ip -6 addr shows local addresses.
Can I get IPv6 if my ISP does not offer it?
A tunnel broker can carry IPv6 over IPv4, but it adds latency and another failure point. For websites, a CDN with IPv6 at the edge is often simpler than changing the origin network.
Is IPv6 secure?
IPv6 is not less secure by design, but it changes the operational model. Every host can have a globally routable address, so firewall rules and host hardening matter. NAT was never a security boundary.
Last updated: . Test your own path with the website checker or read the methodology and about page.