What is IPv6? Addresses, prefixes and proxies
Definition
IPv6 is the version of the Internet Protocol with 128-bit addresses, written as eight groups of up to four hexadecimal digits.
For proxy work, one consequence of IPv6 matters more than the rest: a single customer is handed an enormous block of addresses, so one address on its own stops being a useful identifier and the prefix takes its place. Everything else about filtering, rotation and pool size in IPv6 follows from that.
How an address is laid out
RFC 4291 splits a global unicast address into a routing prefix, a subnet ID and an interface identifier, and requires the interface identifier to be 64 bits for nearly all unicast addresses. Using a documentation address:
2001:0db8:0004 : 0001 : 0000:0000:0000:000a
|-- 48 bits --| 16 b |-------- 64 bits --------|
routing prefix subnet interface ID
(48 here; n bits in RFC 4291)
(the site) (one link) (one host on that link)
written per RFC 5952: 2001:db8:4:1::a
the link it sits on: 2001:db8:4:1::/64
addresses on that link: 18,446,744,073,709,551,616
The short form follows RFC 5952: drop leading zeros in each group, replace the longest run of zero groups with ::, and use lowercase. Python applies the same rules, which is useful when you need to compare addresses that were written differently. On Python 3.15.0:
>>> ipaddress.ip_address("2001:0db8:0000:0000:0000:ff00:0042:8329")
IPv6Address('2001:db8::ff00:42:8329')
>>> ipaddress.ip_address("2001:db8::ff00:42:8329").exploded
'2001:0db8:0000:0000:0000:ff00:0042:8329'
Comparing strings without normalising them first is an easy way to count one address as two.
IPv4 and IPv6 side by side
| IPv4 | IPv6 | |
|---|---|---|
| Address width | 32 bits | 128 bits |
| Written as | Four decimal octets | Eight hex groups, zeros compressed with :: |
| DNS record | A |
AAAA |
| Size of one ordinary subnet | A /24 holds 256 |
A /64 holds 264 |
| What a subscriber usually holds | One address, often shared through carrier-grade NAT | A prefix of at least a /64; RFC 6177 leaves the exact size to operators |
| Stability of one address | Changes when the operator reassigns it | Hosts may rotate temporary addresses on their own (RFC 8981) |
| What a filter can act on in one step | The address, or the block around it | The prefix; blocking a single address costs the sender nothing |
The last row is the one that matters when buying. Because the next request can come from a neighbour in the same allocation, a target that filters IPv6 has to act on a prefix. Which prefix length any particular site uses is not published, and there is no reason two sites would agree, so treat any specific figure you read as a guess unless it comes with evidence. What you can check is how many distinct /64 and /48 blocks your exits actually come from; the subnet entry has a script that groups addresses by /64, and CIDR covers the arithmetic.
Can the target and the proxy even use it?
Two questions decide whether IPv6 exits are useful, and both take seconds to answer. First, does the destination publish an AAAA record at all:
$ dig AAAA example.com +short
2606:4700:10::6814:179a
2606:4700:10::ac42:93f3
That answer was captured on 21 October 2026. A target with no AAAA record cannot be reached over IPv6 by anyone, so an IPv6-only exit is useless against it. Second, does your proxy actually leave over IPv6. The family you use to reach the proxy and the family the exit node uses onward are independent, so a gateway you reach over IPv4 may exit over IPv6, and the reverse. Ask a service that answers on both families what it saw:
curl -s -x http://user:pass@gw.example.net:8000 https://api64.ipify.org
A colon in the answer means the exit reached it over IPv6. The curl guide covers verifying an exit address properly, including why one request is not enough, and residential vs datacenter vs ISP vs mobile lists what else to test before relying on a pool.
Dual-stack means both protocols are available on a machine, not that both are used for a given connection. The client picks, and clients that implement Happy Eyeballs (RFC 8305) start with IPv6 when the destination offers it and fall back to IPv4 if that attempt is slow. Through an explicit proxy that preference only decides how you reach the gateway. The onward choice belongs to the exit, which is why the second test above is the one that answers the question.
Frequently asked questions
Are IPv6 proxies easier or harder to block than IPv4 proxies?
Why do so many proxy products and scraping targets still run on IPv4 only?
Sources
- RFC 8200: Internet Protocol, Version 6 (IPv6) Specification
- RFC 4291: IP Version 6 Addressing Architecture, prefix, subnet ID and 64-bit interface ID
- RFC 5952: A Recommendation for IPv6 Address Text Representation
- RFC 6177: IPv6 Address Assignment to End Sites
- RFC 8981: Temporary Address Extensions for Stateless Address Autoconfiguration in IPv6
- RFC 8305: Happy Eyeballs Version 2, preferring IPv6 with fast fallback
- Python documentation: ipaddress, compressed and exploded forms