You still need pip install numpy --config-settings=--disable-gil . Many C extensions took over a year to become thread-safe. As of today, the PyData stack (NumPy, Pandas, Scikit-learn) is 90% compatible with free-threading, but niche libraries like lxml and cryptography still require the GIL.
| Python Version | % of production downloads | |----------------|---------------------------| | 3.11 | 42% | | 3.12 | 35% | | | 18% | | 3.10 or older | 5% | python 3.13 news today november 2025
It works. The nogil fork is officially dead. For CPU-bound threaded workloads (think web scraping, data processing, or pandas aggregations), teams are reporting 2x to 4x speedups without rewriting a single line of multiprocessing code. the PyData stack (NumPy
— Your friendly local Pythonista