AI codes python 3.12 interpreter in two weeks: discovers true innovation means copying cpython's every dirty little secret

A developer has created a Python 3.12 interpreter, called moonpython, using the Codex AI model, specifically GPT-5.2 and GPT-5.2-Codex, over a period of two weeks in January 2026. The interpreter is designed to run a large, pragmatic subset of Python 3.12, focusing on correctness and a clean architecture, rather than full standard library parity or bytecode compatibility. The project has a test harness with 2,894 tests, including 2,709 generated tests from CPython's Lib/test suite. While the interpreter passes all its current tests, it still has limitations, including missing support for variable annotations, which prevents it from passing the entire CPython Lib/test suite. The developer used Codex to grind through edge cases and iterate until imports behave as expected, demonstrating the potential of AI in handling complex, mundane tasks. The project is available on GitHub, showcasing the power of AI-assisted development in creating complex software projects.