Benchmarks

json-tology is not yet uniformly faster than its comparators. Where we win, where we lose, and where we have work to do.

This page is the honest read on json-tology's performance against four comparators: AJV, Zod, TypeBox (interpreted Value and compiled TypeCompiler), and Valibot. The numbers below come from the bench suite under bench/. Run npm run bench:report to regenerate.

The headline: today, json-tology loses most warm validation scenarios to TypeBox compiled, and loses some warm scenarios to AJV / Valibot. It wins at structural diff, transform encode, schema authoring (extend build), and validating invalid simple data with full error collection. Most of the loss surface comes from the canonical-graph-first execution model: every validate goes through the graph, which buys downstream features (ABox projection, OWL/SHACL emission, semantic round-trip) but pays a per-call cost a code-gen-only validator does not.

We're publishing the full picture rather than cherry-picking. Where we are slow, we say so.

How to read this page

• WIN — json-tology is at least 1.05x faster than the comparator on the listed scenario.
• LOSS — json-tology is at least 1.05x slower.
• EVEN — within 5% either way.
• ops/s — operations per second after warmup. Higher is better.
• ns/op — nanoseconds per operation. Lower is better.
• json-tology vs this — the multiplier between json-tology and the listed library. "2.39x faster" means json-tology runs 2.39 ops for every 1 op of the comparator on this scenario.

The latest run is auto-generated from bench/results/latest.md and is included in full at the end of this page.

What's in scope

The bench suite covers the operations json-tology actually exposes:

Suite	json-tology surface	Comparators
Validation	registry.validate	AJV validate, TypeBox TypeCompiler.Check, Zod safeParse, Valibot safeParse
Instantiation	registry.instantiate (no coercion)	TypeBox Value.Parse, Zod parse, Valibot parse
Coerce	registry.instantiate with castTypes: true, defaults	TypeBox Value.Parse, Zod parse, Valibot parse
Value operations	Value.clone, Value.diff, registry.clean, registry.convert	TypeBox Value.Clean / Convert / Diff, structuredClone
Transforms	Transform.create decode + facade encode	TypeBox Value.Decode / Value.Encode, Zod .transform
Composition	Compose.extend / intersection / discriminatedUnion	TypeBox Type.Composite / Intersect / Union, Zod .extend / intersection / discriminatedUnion, Valibot variant
Serialization	dump, dumpJson, facade encode	JSON.stringify, structuredClone, TypeBox Value.Encode
Registry	cold register + first validate, warm validate	TypeBox TypeCompiler.Compile + Check, Zod, Valibot
Compiled vs Interpreted	SchemaCompiler vs GraphEngine.execute	internal — measures the speedup of the compile path

What's unique

These are operations no comparator implements, so they appear only as a single-library row in the report. They are included for completeness, not as head-to-head wins.

• toTbox — OWL TBox projection from the canonical graph.
• toShacl — SHACL shape projection.
• toQuads / fromQuads — RDF round-trip via projection.
• ABox projection through Materializer.projectAbox.
• findDuplicates over the registry.
• The jt: keyword set (computed properties, invariants, decoders, brands).
• OWL / SHACL emission through OntologyBuilder.

These are surface area that competitors do not expose. They are not wins; they are work the competitors do not do.

Where we have work to do

Pulled directly from bench/results/latest.md. These are scenarios where json-tology is more than 5x slower than the median comparator. Each is a known issue.

• simple valid validation (~6x slower than median) — the per-validate graph traversal cost dominates a 5-property flat schema. Tracked: rework hot path so flat schemas skip subgraph dispatch when no $refs are present.
• nested valid validation (~7x slower than median) — same root cause as simple valid, amplified by per-property subschema lookup. Tracked: precompile a flattened property dispatch table per schema graph.
• convert simple (~32x slower than TypeBox) — castTypes: true runs a separate normalize pass over the value before validate. Tracked: fold normalize into the compiled validator.
• extend + validate cold path (~12x slower than TypeBox) — registering a derived schema rebuilds the canonical graph from scratch per call. Tracked: cache subgraph fragments at register time.
• intersection cold path (~12x slower than Zod) — same root cause as extend + validate.
• dumpJson nested (~8x slower than JSON.stringify) — dump walks the schema graph for every property. Tracked: short-circuit when no Transform encoders are attached anywhere in the subgraph.
• discriminated union warm (~74x slower than TypeBox compiled) — every variant is currently re-resolved through oneOf semantics. Tracked: discriminator-aware fast path that switches directly on the discriminator key.
• cold first validate (~156x slower than Valibot) — Valibot has no compile step at all; the cold path is the warm path. Our cold path includes graph construction, subschema linking, and JIT compilation. Acceptable cost because subsequent calls are fast, but the gap is honest. Tracked: lazy subgraph build for unreachable parts of the schema.

If we are slower on a scenario by more than 5x, you can expect a corresponding tracked item above. None of these are blockers for using the library — but they are real, and we are working on them.

Reproduce

npm install
npm run bench:report

The runner prints to console and writes bench/results/latest.md. Bench numbers move with hardware. The values embedded below came from the developer machine listed in the Environment section. CI runs are uploaded as workflow artifacts (see .github/workflows/bench.yml); use those as the canonical reference once bench-in-CI is established.

For deeper investigation:

npm run bench         # human-readable console output, no markdown
npm run bench:flame   # 0x flame graph profiling under .flame/

Latest run

The block below is auto-generated. To refresh it, run npm run bench:report.

json-tology benchmarks — latest run

Generated by npm run bench:report from bench/run.ts. Numbers move with hardware; treat as a directional snapshot, not absolutes.

Environment

• Date: 2026-05-06T01:04:02.503Z
• Node: v24.13.0
• OS: darwin 25.3.0
• CPU: Apple M4 Max

Scorecard

Comparator	Wins	Ties	Losses	Win rate
typebox	6	0	15	29%
ajv	0	0	3	0%
zod	2	0	11	15%
valibot	0	0	9	0%
structuredClone	1	0	1	50%
JSON.stringify	0	0	1	0%

Validation

simple valid

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	435,362	2297	-	-
typebox	22,388,063	45	51.42x slower	LOSS
ajv	1,596,029	627	3.67x slower	LOSS
zod	550,706	1816	1.26x slower	LOSS
valibot	1,254,756	797	2.88x slower	LOSS

simple invalid

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	240,134	4164	-	-
typebox	293,324	3409	1.22x slower	LOSS
ajv	9,980,953	100	41.56x slower	LOSS
zod	26,308	38011	9.13x faster	WIN
valibot	527,777	1895	2.20x slower	LOSS

nested valid

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	144,609	6915	-	-
typebox	5,565,695	180	38.49x slower	LOSS
ajv	2,817,986	355	19.49x slower	LOSS
zod	197,376	5066	1.36x slower	LOSS
valibot	307,725	3250	2.13x slower	LOSS

Instantiation

instantiate simple

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	240,652	4155	-	-
typebox	415,657	2406	1.73x slower	LOSS
zod	622,687	1606	2.59x slower	LOSS
valibot	1,235,263	810	5.13x slower	LOSS

instantiate nested

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	90,976	10992	-	-
typebox	62,495	16001	1.46x faster	WIN
zod	365,133	2739	4.01x slower	LOSS
valibot	275,905	3624	3.03x slower	LOSS

Coerce

coerce valid

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	336,806	2969	-	-
typebox	217,645	4595	1.55x faster	WIN
zod	3,342,069	299	9.92x slower	LOSS
valibot	2,628,671	380	7.80x slower	LOSS

coerce defaults

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	162,931	6138	-	-
typebox	490,139	2040	3.01x slower	LOSS

Value operations

clean simple

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	174,304	5737	-	-
typebox	407,690	2453	2.34x slower	LOSS

clean nested

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	58,013	17237	-	-
typebox	103,761	9638	1.79x slower	LOSS

convert simple

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	197,050	5075	-	-
typebox	1,889,460	529	9.59x slower	LOSS

clone nested

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	217,000	4608	-	-
structuredClone	150,703	6636	1.44x faster	WIN

diff nested

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	414,935	2410	-	-
typebox	54,124	18476	7.67x faster	WIN

Transforms

decode date

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	269,301	3713	-	-
typebox	3,168,267	316	11.76x slower	LOSS
zod	741,597	1348	2.75x slower	LOSS

encode date

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	1,870,937	534	-	-
typebox	409,693	2441	4.57x faster	WIN

Composition

extend build

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	1,747,352	572	-	-
typebox	705,304	1418	2.48x faster	WIN
zod	12,702	78729	137.57x faster	WIN

extend + validate

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	8,773	113987	-	-
typebox	81,403	12285	9.28x slower	LOSS
zod	9,446	105867	1.08x slower	LOSS

discriminated union

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	396,412	2523	-	-
typebox	50,588,087	20	127.61x slower	LOSS
zod	1,772,771	564	4.47x slower	LOSS
valibot	1,634,634	612	4.12x slower	LOSS

intersection

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	9,907	100940	-	-
typebox	59,743	16738	6.03x slower	LOSS
zod	65,392	15292	6.60x slower	LOSS

Serialization

dump nested

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	46,865	21338	-	-
structuredClone	185,011	5405	3.95x slower	LOSS
typebox	105,605	9469	2.25x slower	LOSS

dumpJson nested

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	44,889	22277	-	-
JSON.stringify	853,520	1172	19.01x slower	LOSS

encode event

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	2,275,822	439	-	-
typebox	955,105	1047	2.38x faster	WIN

Registry

cold first validate

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	2,311	432654	-	-
typebox	19,703	50752	8.53x slower	LOSS
zod	6,271	159471	2.71x slower	LOSS
valibot	295,260	3387	127.76x slower	LOSS

warm validate

Library	ops/s	ns/op	json-tology vs this	Status
json-tology	117,723	8495	-	-
typebox	730,478	1369	6.21x slower	LOSS
zod	1,208,842	827	10.27x slower	LOSS
valibot	292,180	3423	2.48x slower	LOSS

Compiled vs Interpreted

compiled simple valid

Unique to json-tology — no head-to-head comparator.

Library	ops/s	ns/op
compiled	824,433	1213
interpreted	33,288	30041

compiled simple invalid

Unique to json-tology — no head-to-head comparator.

Library	ops/s	ns/op
compiled	509,928	1961
interpreted	34,536	28955

compiled nested valid

Unique to json-tology — no head-to-head comparator.

Library	ops/s	ns/op
compiled	183,375	5453
interpreted	5,785	172859

Where we have work to do

Scenarios where json-tology is more than 5x slower than the median comparator:

• nested valid: 19.49x slower than median comparator
• coerce valid: 7.80x slower than median comparator
• convert simple: 9.59x slower than median comparator
• decode date: 11.76x slower than median comparator
• extend + validate: 9.28x slower than median comparator
• intersection: 6.60x slower than median comparator
• dumpJson nested: 19.01x slower than median comparator
• cold first validate: 8.53x slower than median comparator
• warm validate: 6.21x slower than median comparator

Reproduce

npm install
npm run bench:report

Bench numbers move with hardware. The values committed here came from the developer machine listed in Environment. CI runs are uploaded as workflow artifacts (see .github/workflows/bench.yml).