Benchmarks

Comprehensive performance comparison between Python implementations and the Go version of json2xml.

Test Environment

  • Machine: Apple Silicon (aarch64)

  • OS: macOS

  • Date: January 14, 2026

Implementations Tested

Implementation

Version

Notes

CPython

3.14.2

Homebrew installation

CPython

3.15.0a4

Latest alpha via uv

PyPy

3.10.16

JIT-compiled Python

Go

1.0.0

json2xml-go

Test Data

Size

Description

Bytes

Small

Simple object {"name": "John", "age": 30, "city": "New York"}

47

Medium

bigexample.json (patent data)

2,598

Large

1,000 generated records with nested structures

323,130

Very Large

5,000 generated records with nested structures

1,619,991

Results

Individual Test Results

Test

CPython 3.14.2

CPython 3.15.0a4

PyPy 3.10.16

Go

Small JSON (47 bytes)

75.46ms

55.74ms (1.4x faster)

121.47ms (1.6x slower)

3.69ms (20.4x faster)

Medium JSON (2.6KB)

73.87ms

57.98ms (1.3x faster)

125.73ms (1.7x slower)

4.32ms (17.1x faster)

Large JSON (323KB)

419.67ms

328.98ms (1.3x faster)

517.51ms (1.2x slower)

67.13ms (6.3x faster)

Very Large JSON (1.6MB)

2.09s

1.86s (1.1x faster)

1.42s (1.5x faster)

287.58ms (7.3x faster)

Summary (Average Across All Tests)

Implementation

Avg Time

vs CPython 3.14.2

Go

90.68ms

7.34x faster 🚀

PyPy 3.10.16

545.58ms

1.22x faster

CPython 3.15.0a4

575.45ms

1.16x faster

CPython 3.14.2

665.23ms

baseline

Key Observations

1. Go is the Clear Winner

Go outperforms all Python implementations by a significant margin:

  • 7.34x faster than CPython 3.14.2 on average

  • Up to 20x faster for small inputs due to minimal startup overhead

  • Consistent performance across all input sizes

2. CPython 3.15.0a4 Shows Promising Improvements

The latest Python alpha demonstrates consistent performance gains:

  • 13-35% faster than CPython 3.14.2 across all test sizes

  • Improvements likely due to ongoing interpreter optimizations

3. PyPy Has Interesting Trade-offs

PyPy’s JIT compiler creates a unique performance profile:

  • Slower for small/medium inputs: JIT compilation overhead hurts for quick operations

  • Faster for very large inputs: JIT shines on the 5K record test (1.5x faster than CPython)

  • Best suited for long-running processes or batch processing

4. Startup Overhead Dominates Small Inputs

Python’s interpreter startup time is significant:

  • CPython takes 55-75ms even for 47 bytes of JSON

  • Go takes only 3.7ms for the same operation

  • For CLI tools processing small files, Go provides a much better user experience

When to Use Each Implementation

Use Case

Recommended

CLI tool for small/medium files

Go (json2xml-go)

High-throughput batch processing

Go or PyPy

Integration with Python codebase

CPython 3.15+

One-off conversions in scripts

CPython (any version)

Running the Benchmarks

Python Multi-Implementation Benchmark

# Set the Go CLI path
export JSON2XML_GO_CLI=/path/to/json2xml-go

# Run the benchmark
python benchmark_multi_python.py

Simple Python vs Go Benchmark

# Set paths via environment variables (optional)
export JSON2XML_GO_CLI=/path/to/json2xml-go
export JSON2XML_EXAMPLES_DIR=/path/to/examples

# Run the benchmark
python benchmark.py

Reproducing Results

  1. Install required Python versions using uv:

    uv python install 3.14 3.15.0a4 pypy@3.10

  2. Build the Go binary:

    cd /path/to/json2xml-go
go build -o json2xml-go ./cmd/json2xml-go

  3. Run the multi-Python benchmark:

    cd /path/to/json2xml
python benchmark_multi_python.py