Skip to content

Compare 200+ serialization libraries across 14 languages with fair within-language rankings—not marketing microbenchmarks.

Open live Dashboard Start Serialization 101 How we measure


What this project is

This suite runs the same data shapes through serializers in many programming languages, writes a shared CSV contract, and analyzes results with one statistics pipeline. Use it to compare libraries inside one language, one standard, one data set, and one data type, not to crown a single global winner.

You want… Go here
Interactive charts, Pareto trade-offs, compare lab Dashboard
One-question tests (why we do not only use the big table) Experiments
Does this library match the RFC / spec section? Dashboard · Compliance (#compliance/{lang}/{serializer}) · catalog notes
Course on formats and trade-offs (101–401) Learn
Per-language runtime primer, library lists, and caveats Languages (sidebar: C, C#, C++, …)
Methodology, metrics, how to add a codec Benchmarks
The five data sets (suite, columnar, graph, array, tensor) Data sets
The words this site uses (data set, data type, runner) Terminology

Same data · same pipeline

The live Dashboard is the published home for measured numbers. Pick a language, a standard, and a data type, then read speed vs size and the ranking. Language indexes list the roster and caveats.

Live Dashboard: language and data-type filters, speed vs size scatter, and throughput ranking


Languages

Fourteen runners share the same data types and the same analysis rules.

Language Serializers
C · Dashboard 20
C# · Dashboard 50
C++ · Dashboard 35
Fortran · Dashboard 9
Go · Dashboard 26
Java · Dashboard 27
JavaScript · Dashboard 25
Kotlin · Dashboard 32
Mojo · Dashboard 13
PHP · Dashboard 15
Python · Dashboard 23
Rust · Dashboard 22
Swift · Dashboard 14
Zig · Dashboard 17

C++ is 35 when ARROW_ROOT points at Arrow 25.0.1. Without that prefix the five Arrow rows are skipped and sbe still registers.


Learn the ideas first (optional path)

If you are new to serialization, start with the course under Learn:

  1. Serialization 101 — workload, standard, implementation, and evaluation
  2. Serialization 201 — how formats work under the hood
  3. Serialization 301 — production judgment
  4. Serialization 401 — wire formats and labs

Then return to the Dashboard with clearer questions.


Honesty rules (short)

  • Prefer comparisons within one language, one standard, one data set, and one data type. A family is a teaching cut, coarser than the standard. The sets are on Data sets. See Families.
  • Implementation quality often matters more than the format brand name.
  • Payload shape changes costs a great deal.
  • Numbers on this site are from this suite’s runners and analysis—not a universal ranking of all software.

Full methodology: Benchmarks overview · Analysis methodology · Claims and replication.