Py2Cpp Accelerator

Convert Python code to optimized C++ using frontier LLMs, achieving up to 60,000x performance gains.

How it works

1. You provide Python code (or use the built-in pi approximation example)
2. An LLM (OpenAI, Anthropic, Gemini, or Grok) generates equivalent, highly optimized C++
3. The C++ is compiled with aggressive optimization flags and executed
4. You see the speedup

Requirements

• Python 3.11+
• uv
• A C++ compiler: clang++, g++, or MSVC cl (auto-detected, see Compiler setup below)

Setup

git clone https://github.com/anton-karlovskiy/py2cpp-accelerator.git
cd py2cpp-accelerator
cp .env.example .env   # fill in your API keys
uv sync

API keys

Edit .env with the keys for the models you want to use:

OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_API_KEY=...
GROK_API_KEY=xai-...

Only the key for your chosen --model is required.

Usage

# Run the built-in pi example with OpenAI (default)
uv run python main.py

# Show Python baseline timing alongside the C++ result
uv run python main.py --python-baseline

# Use a specific model
uv run python main.py --model anthropic
uv run python main.py --model gemini
uv run python main.py --model grok

# Race all four models against each other
uv run python main.py --model all --python-baseline

# Convert your own Python file
uv run python main.py --input samples/my_script.py --model openai

# Generate C++ only — skip compile and run
uv run python main.py --input samples/my_script.py --no-compile

Models

Flag	Provider	Model
openai	OpenAI	gpt-5
anthropic	Anthropic	claude-sonnet-4-6
gemini	Google	gemini-2.5-pro
grok	xAI	grok-4

Performance

The built-in example approximates π using 200 million iterations of the Leibniz formula. Typical results on the same hardware:

Implementation	Time	Speedup
Python	~19 s	1×
C++ (best)	~0.013 s	~1,450×

With vectorization and -Ofast, the best models have demonstrated speedups exceeding 60,000× on compute-bound loops.

Compiler setup

The compiler is auto-detected at runtime. You just need one of the following on your PATH:

Compiler	Windows install	macOS/Linux install
clang++	winget install LLVM.LLVM	brew install llvm / apt install clang
g++	Install via MSYS2	brew install gcc / apt install g++
MSVC cl	Visual Studio Build Tools	—

To verify what's detected on your machine:

uv run python system_info.py

Project structure

py2cpp-accelerator/
├── main.py                        # CLI entrypoint
├── system_info.py                 # detects OS, CPU, and available compilers
├── samples/
│   └── pi_approximation.py        # built-in π benchmark (Leibniz formula, 200M iterations)
├── pyproject.toml
├── .env.example
└── .gitignore