Python 3.13.1 Released November 2025 Instant

By Christmas Eve, the community had stabilized. The PyPI daily download count for packages marked python_version >= '3.13.1' had tripled. NumPy released 2.5.0 with explicit subinterpreter support, yielding 4x speedups on large matrix multiplications. Even Django’s ASGI server got a patch that let each request handler spin up in its own lightweight subinterpreter, wiping out the last of the global connection-pool bottlenecks.

A critical CVE was announced—a use-after-free in the new biased reference counter, only triggerable when mixing subinterpreters and C extensions that manually manipulated PyObject* refcounts. The entire Python security team held an emergency sprint. python 3.13.1 released november 2025

By midnight, she had a test. A stupid test, the kind you write when you’re too tired to be clever: a million iterations of a mandelbrot calculation, split across four subinterpreters, each chewing on a different quadrant of the complex plane. By Christmas Eve, the community had stabilized

futures = [] with interp1.run() as i1: futures.append(i1.call(mandel_chunk, (-2.0, -1.0, -1.5, 0.0, 500, 500))) # ... repeat for four interpreters ... Even Django’s ASGI server got a patch that

Kavya’s closing slide read: “Python isn’t slow anymore. You just haven’t updated.”

- match now supports case with guard: as a native keyword expression. No more parentheses gymnastics. Elena leaned back, her chair creaking. The subinterpreters were the real story. For years, Python had been a single-threaded soul trapped in a multi-core world. You could spawn processes, but they were heavy. You could use asyncio , but it was cooperative. True parallelism—without the GIL’s chaperone—had always been the dream deferred.