Definition

Random bits are the portion of a UUID filled by a Cryptographically Secure Pseudo-Random Number Generator (CSPRNG). They are the primary source of uniqueness — the higher the count of random bits, the lower the probability that two independently generated UUIDs are identical.

UUID bit layout comparison

UUID v4 — 122 random bits:
xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx
← rand  →     ↑v4  ↑var ←  rand    →
  All x bits are random (minus version + variant = 6 fixed bits)

UUID v7 — 74 random bits:
xxxxxxxx-xxxx-7xxx-yxxx-xxxxxxxxxxxx
←48-bit ms→   ↑v7  ↑var←rand_a→←rand_b→
               12 bits    62 bits random

CSPRNG sources per platform

PlatformCSPRNG source
Node.jscrypto.randomFillSync()
Browsercrypto.getRandomValues()
Pythonos.urandom()
Gocrypto/rand.Read()
Rustgetrandom crate
JavaSecureRandom

All modern UUID libraries use CSPRNG sources. Never use Math.random() or equivalent non-cryptographic PRNGs for UUID generation — they produce predictable values with far fewer effective random bits. See CSPRNG.

Random bits and collision probability

The relationship between random bits b and the number of UUIDs n needed for a 50% collision probability follows the birthday problem approximation:

n ≈ √(2 · 2^b · ln 2)
UUID versionRandom bits~50% collision at
UUID v41222.7 × 10¹⁸ values
UUID v7746.1 × 10¹¹ values per millisecond
UUID v7 (with monotonic counter)~62 effectiveCollisions eliminated within one generator

At one billion UUIDs per second, UUID v4 would take 85 years to reach a 50% collision probability. UUID v7 provides weaker per-millisecond guarantees but the monotonic counter in RFC 9562-compliant implementations eliminates same-millisecond collisions within a single generator. See UUID Collision.

UUID v7 rand_a as monotonic counter

RFC 9562 §6.2 recommends Method 1 for same-millisecond ordering: use some of the rand_a bits as a monotonic counter that increments for each UUID generated within the same millisecond.

UUID v7 with monotonic counter (Method 1):
[48-bit ms][4-bit ver][12-bit counter][2-bit var][62-bit random]
                       ↑ increments each UUID within same ms

The counter starts at a random value on each new millisecond, preserving sufficient randomness while guaranteeing strict ordering within a millisecond on a single generator instance. The 62-bit rand_b remains purely random.

Frequently asked questions

How many random bits does UUID v4 have?

UUID v4 has 122 random bits. Of the 128 total bits, 6 are fixed: 4 bits for the version field (0100) and 2 bits for the variant field (10). The remaining 122 bits are filled by a CSPRNG. This gives approximately 5.3 × 10³⁶ possible values.

How many random bits does UUID v7 have?

UUID v7 has 74 random bits: 12 bits in the rand_a field and 62 bits in the rand_b field. The remaining 54 bits are: 48-bit Unix millisecond timestamp, 4-bit version (0111), and 2-bit variant (10). Implementations may use some rand_a bits as a monotonic counter (per RFC 9562 §6.2), reducing effective randomness slightly but eliminating same-millisecond collisions entirely.

Is 74 random bits enough for UUID v7 to be collision-safe?

Yes, for any realistic workload. With 74 random bits, you need to generate more than 100 billion UUIDs in the same millisecond before a collision becomes likely. No production system approaches this rate. See UUID Collision for the birthday problem calculation.