Benchmarks

schiva compared with ajv and nine other JSON Schema validators: how many tests of the JSON-Schema-Test-Suite each one passes, how fast it validates the suite and realistic payloads, how fast it compiles a schema, and how fast the features beyond validating are.

Summary

Every chart compares libraries doing the same work: pick the mode above it. true/false only tells whether the value is valid; first error stops at the first failing check and builds its error; all errors builds every error. Longer is faster, and next to each library is how many times slower than schiva it is.

Tests passed

Tests of the JSON-Schema-Test-Suite each library gets right, draft by draft (a schema it cannot compile counts as wrong for every test of its group). Only schiva and ajv run draft-04 and draft-06 here; drafts 2019-09 and 2020-12 run with the libraries that implement them.

Speed on the test suite

How many times per second each library validates every test of the groups all the libraries of the draft pass, so that each one does the same work.

ajv's speed on drafts 2019-09 and 2020-12 changes a lot from one process to the next (between about 4k and 15k runs per second in our runs).

Speed on payloads

Validations per second of one value, valid and invalid, by each library that compiles the schema and gives the right answer for it. The schemas are in bench/lib/cases.js.

Compiling

Schemas per second, from the order schema to a function that validates it. jsonschema, tv4 and @cfworker/json-schema interpret the schema instead of compiling it: their "compiling" only builds an object, and they pay for it at every validation, which is why they are the slowest on the payloads above.

Speed of the features

schiva and ajv on the features beyond validating JSON Schema, each with the same options and plugins (ajv-keywords, ajv-formats, ajv-draft-04, ajv's standalone code), with schiva's speed in times ajv's. With the options that change the data, each call validates a new object, made the same way for both. The cases are in bench/features.js.

schiva and ajv, feature by feature

See Migrating from ajv for the options, errors and APIs of each one.

Libraries

Every library of json-schema-benchmark that runs on current Node.js, with the drafts and modes each one is benchmarked in. (ajv always returns true or false, and builds the first error, or all of them, on the way.)

Method

• Every measurement runs in a new Node.js process with a single library, as a service uses it, so libraries cannot slow each other down through the state of the JIT they would share. Each one is timed for 1 second after a warm-up, in 3 processes, and the median is kept.
• Measured on 13th Gen Intel(R) Core(TM) i7-13700H, 20 cores, Windows_NT, Node.js v22.21.1, on 2026-09-30. Benchmarks depend on the machine and on what else runs on it: compare the libraries with each other, not these numbers with another machine's.

cd bench
pnpm install
pnpm start                          # test suite and payloads, isolated (about 30 minutes)
pnpm run features                   # the features of schiva and ajv
pnpm run conformance draft2020-12   # tests passed by schiva and ajv, file by file
pnpm run report                     # writes this page