A different reason to subnet
In IPv4, subnetting is mostly about conservation: addresses are scarce, so you carve a block into the smallest pieces that still fit each network (the logic in the subnetting basics article). IPv6 removes that pressure entirely. A single /64 subnet contains 2^64 addresses, more than the entire IPv4 internet squared, so you never subnet an IPv6 network to save space. You subnet it to give structure: one prefix per VLAN, per site, per customer. The math is the same prefix arithmetic as IPv4 (the tool computes it with arbitrary-precision integers), but the mindset is the opposite.
Why a subnet is a /64
The single most important rule in IPv6 addressing is that a normal subnet is a /64, leaving 64 bits for the interface identifier. This is not a suggestion; large parts of the protocol assume it. Stateless address autoconfiguration (covered in the address configuration article) builds a host's address by appending a 64-bit interface identifier to the announced prefix, and the modified EUI-64 derivation that the tool can perform produces exactly 64 bits. Use a longer prefix like a /80 or /112 on a link and autoconfiguration simply stops working. The addressing architecture (RFC 4291) is built around this boundary, so the practical advice is blunt: make every end-user subnet a /64 and do not try to be frugal inside it.
The allocation hierarchy
If every subnet is a /64, sites and providers work in larger blocks above that:
/48 a site 65,536 subnets (/64s)
/56 a small site 256 subnets (/64s)
/64 one subnet the link itself
An ISP typically delegates a /48 or /56 to a customer, who then assigns /64s out of it to individual links. RFC 6177 walks back the original "/48 for everyone" guidance toward giving each site enough space to subnet comfortably, which in practice means a /48 or /56. The point is that you have room to spare: even a /56 gives a home or small office 256 separate networks, far more than it will ever use.
Subnet on nibble boundaries
Because IPv6 is written in hex and each hex digit is 4 bits (a nibble), prefixes that fall on a 4-bit boundary are far easier to read and manage. A /48, /52, /56, /60, and /64 all end cleanly between hex digits, so the subnet portion lines up with whole characters in the address. Subnetting a /48 into /52s gives you 16 blocks distinguished by a single hex digit; going to /56 gives 256 blocks across two digits. Staying on nibble boundaries keeps your addressing plan legible, which matters more in IPv6 than squeezing out the last bit ever could.
Counting subnets and addresses
The arithmetic is straightforward. The number of /64 subnets inside a shorter prefix is 2 raised to the difference in prefix lengths: a /48 holds 2^(64-48) = 2^16 = 65,536 subnets. The number of addresses inside any /64 is 2^(128-64) = 2^64. The tool reports both the address count for a prefix and the usable range, and the numbers get large quickly, which is exactly the point: scarcity is not a constraint you design around anymore.
The takeaway
IPv6 subnetting flips the IPv4 habit on its head. You are not dividing a scarce resource; you are imposing structure on an effectively unlimited one. Give every subnet a /64 so autoconfiguration and neighbor discovery work as designed, request a /48 or /56 so you have room to organize, and subnet on nibble boundaries so your plan stays readable. When you paste a prefix into the tool and see the address count, treat that abundance as permission to design for clarity rather than for conservation.