Check what your own connection does on /ipv6-test, which tests each protocol separately and reports both addresses.
The differences that actually matter
IPv4 is defined by RFC 791, published in 1981. IPv6 is defined by RFC 8200, the 2017 standard that replaced RFC 2460.
| IPv4 | IPv6 | |
|---|---|---|
| Address length | 32 bits | 128 bits |
| Total addresses | 4,294,967,296 | 340 undecillion (2^128) |
| Notation | 203.0.113.42 |
2001:db8::a1b2 |
| Typical home allocation | 1 shared address | a /56 or /64 prefix |
| Header size | 20 bytes, variable | 40 bytes, fixed |
| Header checksum | Yes | Removed, left to lower layers |
| Fragmentation | Routers may fragment | Source only, using Path MTU Discovery |
| Address assignment | DHCP or static | SLAAC, DHCPv6, or static |
| Broadcast | Yes | Replaced by multicast |
| NAT | Near universal | Rare and discouraged |
The fixed 40-byte header is the reason IPv6 forwarding is cheaper for routers even though the addresses are four times longer. Options moved into extension headers, so a router reads a predictable block and moves on.
How addresses are written
An IPv6 address is eight groups of four hexadecimal digits. RFC 5952 sets the rules for shortening it: drop leading zeros in each group, and replace one run of all-zero groups with ::.
2001:0db8:0000:0000:0000:0000:0000:0001
2001:db8::1
The :: shortcut appears at most once, because two of them would be ambiguous. The first 64 bits are normally the network prefix your ISP routes to you, and the last 64 bits identify the interface.
Prefixes you meet in practice: 2000::/3 is the global unicast space, fe80::/10 is link-local and present on every interface, and fd00::/8 is the unique-local range from RFC 4193.
NAT is the real difference for a home network
IPv4 ran out of room, so ISPs give a household one public address and the router translates every internal connection onto it. Inbound connections have nowhere to land unless you configure a port forward. Some ISPs go further and put your router behind carrier-grade NAT, where you do not hold a public IPv4 address at all.
IPv6 gives your router a whole prefix, so every device gets its own globally routable address. Nothing is translated. Inbound traffic is controlled by the firewall rule set rather than by address scarcity, which is a policy decision instead of an accident of addressing.
Two consequences follow. Peer-to-peer software, game hosting, and self-hosted services are simpler over IPv6. And the address on your laptop is the address a website records, which is why RFC 8981 privacy addresses exist: the operating system rotates the interface half of the address so the same device does not stay identifiable across sites.
Dual stack, and how the browser chooses
Almost nothing runs IPv6 alone. Devices run both, which the industry calls dual stack. When you open a hostname with both an A record and an AAAA record, the browser starts an IPv6 connection, waits a short interval, then starts an IPv4 connection in parallel and keeps whichever completes first. That algorithm is Happy Eyeballs version 2, RFC 8305.
The practical effect is that broken IPv6 rarely produces a visible failure. It produces a delay of a few hundred milliseconds, and then IPv4 quietly wins. Our IPv6 test works around that by using hostnames that publish only one record type, so each protocol is measured on its own.
Adoption, measured rather than predicted
Google publishes a daily measurement of the share of its users that reach it over IPv6 on its IPv6 statistics page.
| Date | IPv6 share of users reaching Google |
|---|---|
| January 2016 | 8 to 10% |
| January 2020 | 26 to 31% |
| January 2023 | 38 to 43% |
| January 2026 | 44 to 50% |
| August 2026 | around 46% |
The ranges are real, not rounding. Weekend numbers run several points above weekdays because home connections carry more IPv6 than corporate ones. Country figures vary far more than the global average: India, France, Germany, and Malaysia sit well above it, while large parts of Africa and Central Asia sit in single digits.
Adoption climbed roughly four points a year over the past decade. At that rate IPv4 stays load-bearing into the 2030s, which is why every serious deployment still tests both.
What to check on your own connection
Open /ipv6-test and read three fields: whether an IPv6 address exists, whether it is global rather than link-local, and whether the ASN behind your IPv6 address matches the one behind your IPv4 address. A mismatch usually means a tunnel or a VPN is carrying one protocol and not the other.
From a terminal:
curl -4 https://v4.whatsmyip.fyi/ip
curl -6 https://v6.whatsmyip.fyi/ip
If the second command fails with “Could not resolve host” or “Network is unreachable”, your network has no working IPv6 path. That is a router or ISP setting, not a fault on your device.
Questions people ask
- Is IPv6 faster than IPv4?
- Usually the same, sometimes slightly faster. IPv6 removes NAT from the path and often takes a cleaner route to large content networks. The gain is in the low single digits of milliseconds, not something you would notice while browsing.
- Should I disable IPv6 on my router?
- No. Disabling it forces every connection through your ISP's IPv4 path, which is often carrier-grade NAT. If a specific application breaks, fix that application. Windows, macOS, iOS, and Android all prefer IPv6 and fall back to IPv4 on their own.
- Will IPv4 be switched off?
- Not on any announced date. The registries ran out of free IPv4 blocks between 2011 and 2019, but existing addresses keep working and get traded on a secondary market. Plan for both protocols indefinitely.
Related
Last reviewed 2026-09-04. Reviewed quarterly, or sooner when a vendor changes something.