Fast, zero-dependency SemVer for ESM and TypeScript, with functional, tree-shakeable APIs.
- β Complete SemVer version and range toolkit.
- π Faster than node-semver across tested operations.
- π¦ Pure ESM with zero runtime dependencies.
- π First-class TypeScript declarations.
- π³ Functional, tree-shakeable named exports.
- π Mutable
SemVerandSemVerRangerecords. - β‘ 26.0% smaller for full CDN imports.
- πͺΆ 60.1% smaller with common bundled imports.
- π‘οΈ Immutable collection operations.
npm add verkitimport {
coerce,
increment,
normalize,
normalizeFull,
parse,
truncate,
} from 'verkit'
const version = parse('1.2.3-rc.1+sha.abc')
version.patch = 4
normalizeFull(version) // '1.2.4-rc.1+sha.abc'
normalize(version) // '1.2.4-rc.1'
increment(version, 'minor') // '1.3.0'
truncate(version, 'patch') // '1.2.4'
coerce('release 42.6.7.9', { rtl: true }) // '6.7.9'Version APIs accept strings or mutable SemVer objects returned by parse.
normalizeFull keeps build metadata; normalized, incremented, and truncated
versions omit it.
import { compare, compareBuild, sortReversed } from 'verkit'
compare('1.0.0+one', '1.0.0+two') // 0
compareBuild('1.0.0+one', '1.0.0+two') // -1
sortReversed(['1.0.0', '2.0.0']) // ['2.0.0', '1.0.0']compare ignores build metadata; compareBuild uses it as a tie-breaker.
import {
findMaxSatisfying,
normalizeRange,
parseRange,
satisfies,
} from 'verkit'
const range = parseRange('^1.2.3')
normalizeRange(range) // '>=1.2.3 <2.0.0-0'
satisfies('1.5.0', range) // true
findMaxSatisfying(['1.2.3', '1.5.0', '2.0.0'], range) // '1.5.0'Range APIs accept strings or mutable SemVerRange objects. They support
comparators, unions, hyphens, wildcards, tilde, caret, loose parsing, and
prereleases.
See the API reference.
parse and parseRange throw detailed TypeErrors. Their safe wrappers,
tryParse and tryParseRange, return null; other safe transforms and
predicates keep their documented null/false behavior.
Only renamed or reshaped node-semver APIs are listed; same-named functions
such as clean, coerce, compare, and satisfies are omitted.
| node-semver | verkit |
|---|---|
SemVer |
parse |
parse |
tryParse |
valid |
normalize |
inc, diff |
increment, difference |
major, minor, patch, prerelease |
getMajor, getMinor, getPatch, getPrerelease |
rcompare, compareLoose, cmp |
compareReversed, compare with { loose: true }, compareWithOperator |
eq, neq, gt, gte, lt, lte |
isEqual, isNotEqual, isGreater, isGreaterOrEqual, isLess, isLessOrEqual |
rsort |
sortReversed |
rcompareIdentifiers |
compareIdentifiersReversed |
Comparator |
SemVerComparator, normalizeComparator, satisfiesComparator, comparatorsIntersect |
Range |
parseRange |
toComparators, validRange |
rangeToComparators, normalizeRange |
maxSatisfying, minSatisfying, minVersion |
findMaxSatisfying, findMinSatisfying, findMinimumForRange |
outside, gtr, ltr |
isOutsideRange, isGreaterThanRange, isLessThanRange |
intersects, subset |
rangesIntersect, isRangeSubset |
RELEASE_TYPES |
INCREMENT_TYPES (also includes release) |
valid returns a normalized string | null in node-semver, so its equivalent
is normalize. Use isValid when you only need a boolean.
Use options objects such as { loose: true } and { identifier, identifierBase }.
verkit follows node-semver semantics with three user-visible differences:
- Array helpers never mutate their inputs.
- verkit is ESM-only, with no CommonJS, CLI, or
NODE_DEBUG=semveroutput. - Error text, stack traces, and supported runtimes may differ.
Full package imports, minified with Rolldown:
| Package | Minified | gzip | Brotli |
|---|---|---|---|
| verkit | 18,232 B | 5,730 B | 5,215 B |
| semver | 24,648 B | 7,361 B | 6,708 B |
| verkit reduction | 26.0% | 22.2% | 22.3% |
Common validation, range, comparison, increment, and coercion imports, tree-shaken and minified with Rolldown:
| Package | Minified | gzip | Brotli |
|---|---|---|---|
| verkit | 9,898 B | 3,384 B | 3,084 B |
| semver | 24,801 B | 7,432 B | 6,773 B |
| verkit reduction | 60.1% | 54.5% | 54.5% |
Run pnpm test:size to reproduce the comparison.
Measured on a MacBook Pro with an Apple M1 Max and 32 GB RAM. Higher is better.
| Operation | verkit ops/s | semver ops/s | Faster |
|---|---|---|---|
| Parse and normalize | 1.92M | 1.63M | verkit 1.18Γ |
| Compare | 1.60M | 1.13M | verkit 1.41Γ |
| Compare parsed versions | 12.83M | 5.93M | verkit 2.16Γ |
| Increment | 1.78M | 1.02M | verkit 1.76Γ |
| Coerce | 1.18M | 1.05M | verkit 1.13Γ |
| Satisfy uncached ranges | 0.07M | 0.06M | verkit 1.18Γ |
| Satisfy pre-parsed inputs | 5.15M | 2.28M | verkit 2.26Γ |
Range benchmarks either cycle through 1,001 inputs to avoid cache hits or parse once and reuse the resulting objects.
Run runtime benchmarks with pnpm bench.
MIT Β© 2026-PRESENT Kevin Deng.
Parts of the implementation and test fixtures are derived from node-semver under the ISC license; see THIRD_PARTY_NOTICES.md.