IPv6 Address Type Identifier

The Three Core Types of IPv6 Addresses: Unicast, Multicast, and Anycast

Every IPv6 address category ultimately resolves to one of three delivery models. Each is a network layer identifier with a different fan-out behavior:

• Unicast addresses — identify a single interface. Packets sent to a unicast address are delivered to exactly one destination: a one-to-one relationship.
• Multicast addresses — identify a set of interfaces that have joined the same group. A packet sent to a multicast address is delivered to every member: one-to-many.
• Anycast addresses — identify a set of interfaces, but delivery goes to whichever one is topologically nearest according to the routing protocol: one-to-nearest.

Notice what's missing: broadcast. IPv6 has no broadcast address category at all — the role broadcast played in IPv4 networking is handled entirely by multicast in IPv6, most often through the all-nodes group.

Global Unicast Addresses: The Public IPv6 Address Format

Global unicast addresses are the IPv6 equivalent of public IPv4 addresses — routable across the internet and assigned from the block 2000::/3. IANA allocates chunks of this IPv6 address space to a Regional Internet Registry such as ARIN, which then delegates smaller blocks down to ISPs and, eventually, individual sites, each carrying its own routing prefix. The structure is aggregatable by design: a 48-bit global routing prefix — a network prefix, in IANA's own terms — identifies the site, a 16-bit subnet ID (sometimes called the Site-Level Aggregator) lets that site carve out internal subnets, and the remaining bits form the interface identifier. Because every hexadecimal digit encodes four bits, this aggregatable global unicast structure is easy to fingerprint — in practice, essentially every IPv6 address seen on the public internet today starts with the digit 2 or 3.

Link-Local Addresses: Automatic and Non-Routable

Link-local addresses use the prefix FE80::/10 and are auto-assigned to every IPv6 network interface the moment it comes up, whether or not a router is present. The host portion is generated either from a modified EUI-64 derived from the interface's MAC address, or — on most modern IPv6 stacks — from a randomized value chosen for privacy. IPv6 routers never forward packets with a link-local source or destination beyond their originating segment, which makes this address family the backbone of neighbor discovery and address autoconfiguration rather than general-purpose IPv6 routing.

Unique Local Addresses: IPv6's Private Address Range

Unique local addresses play the same role that private address ranges (like 10.0.0.0/8) play in IPv4: internal-only communication that never touches the internet protocol backbone. The block is FC00::/7, and in practice almost every IPv6 deployment sets the eighth "L" bit to 1, producing prefixes beginning with FD00::/8. A 40-bit pseudorandom global ID, written as ten hexadecimal characters, is appended to that prefix, which makes accidental collisions between two merging IPv6 networks statistically negligible without needing an ISP-independent registration authority. Because the prefix is self-generated rather than provider-assigned, these networks stay stable even if you switch providers, and unlike a public address, they are filtered at IPv6 site boundaries by policy rather than by NAT alone — though some deployments occasionally still lean on NAT66 or CIDR-based filtering at the edge for extra isolation.

Loopback and Unspecified Addresses

Two special single-address categories round out the unicast family. The loopback address, ::1/128, is the 128-bit equivalent of IPv4's 127.0.0.1 and is used for testing the local TCP/IP stack via localhost — packets sent to this loopback interface are returned straight back on the same IPv6 interface and never touch the wire. The unspecified address, ::/128, is the counterpart to 0.0.0.0 and is used only as a placeholder — by a DHCP client before it has a real IPv6 address, for instance — and must never be assigned to a network node or forwarded by an IPv6 router. Any router that receives a packet sourced from the unspecified address should silently drop it, the same defensive rule that applies to the loopback address.

Embedded IPv4-in-IPv6 and Transitional Addresses

During the long transition from IPv4, several address formats exist purely to bridge the two protocols. An embedded IPv4 address takes an IPv4 address A.B.C.D and maps it into the low-order bits of an IPv6 address as ::A:B:C:D, a pattern generically called IPv4-in-IPv6. A closely related but distinct canonical address form, the IPv4-mapped address (::ffff:0:0/96), lets a dual-stack application represent an IPv4 peer using IPv6 sockets. Teredo (2001::/32) is a further transition mechanism that runs an automatic tunnel for IPv6 traffic across an IPv4-only NAT, encoding the client's public IPv4 address directly inside the IPv6 address itself.

Anycast Addresses: The One-to-Nearest IPv6 Address Category

Anycast addresses have no reserved prefix of their own — they're carved out of ordinary global unicast address space and then deliberately configured on more than one interface, which is what makes an anycast address indistinguishable from a regular one just by looking at it. Configuring that same anycast address on several IPv6 routers at once is what creates the effect in the first place, and once multiple devices advertise the address into the routing table, the routing protocol's own metric decides which nearest server actually receives each packet, which is exactly why this address category is popular for load balancing across geographically spread DNS resolvers or CDN nodes without a dedicated appliance in front of them.

IPv6 Addressing in Practice: Autoconfiguration and Neighbor Discovery

Classification isn't just an exercise — it drives how an IPv6 host actually configures itself. Stateless address autoconfiguration (SLAAC) lets a host build its own global address by combining a router-advertised prefix with a locally generated host portion, with no central server involved. Stateful DHCPv6, by contrast, hands out addresses — and other configuration, like DNS resolver addresses — from a central pool, the same way DHCP does for IPv4. Neighbor discovery, IPv6's replacement for ARP and ICMP router discovery, leans on these address families at every step: IPv6 routers advertise prefixes to the all-nodes address, hosts verify uniqueness via duplicate address detection against a solicited-node group, and every exchange happens over link-local addresses regardless of which global address is being configured.

A few address types exist purely to manage the address lifetime of a SLAAC-generated address for privacy — a concern the Internet Engineering Task Force addressed directly in the RFC that defines temporary addressing. A temporary address is periodically regenerated and eventually deprecated so an IPv6 host isn't trivially trackable across networks by a fixed host identifier; the underlying method for choosing that identifier — whether a cryptographically generated address tied to a public/private key pair, or the simpler stable privacy address scheme — determines how predictable it is to an outside observer. When a host has several valid addresses of different families and scopes, an algorithm called default address selection decides which one to actually use as the source for a new connection, and applications resolving a hostname through the Domain Name System will typically prefer whatever a DNS lookup returns as the most global address available over anything scoped more narrowly.

IPv6 vs IPv4 Address Types: What Changed and Why

Anyone coming from IPv4 networking will recognize most of these categories as familiar ideas with far more room to breathe. IPv4 packed its addressing into 32-bit values, giving roughly 4.29 billion possible addresses — a number that felt enormous decades ago and is now a genuine constraint on global IPv4 routing. IPv6's 128-bit space isn't just bigger for its own sake: it's what makes it practical to give every IPv6 address family its own dedicated, unambiguous prefix rather than relying on classful conventions, VLSM tricks, or manual workarounds to stretch a scarce resource.

A few concrete differences stand out. Broadcast disappears entirely, replaced by multicast. Private IPv6 addressing moves from provider-dependent IPv4 ranges to unique local addresses that are globally unique by construction, even without a central registry. And because every network node auto-configures a link-local address the instant it initializes, basic on-link communication and address resolution work even before a router or any address server gets involved. The packet header itself was simplified too, dropping several rarely used IPv4 fields so IPv6 routers can walk the routing table faster for any address category.