PyTorch adds Python 3.15 support to nightly wheel builds

PyTorch's latest pull request (#184600) extends nightly wheel builds to include Python 3.15 and Python 3.15t (the free-threaded, or “nogil” variant) for Linux manywheel targets. Authored by Claude and approved by core maintainers malfet and albanD, the change adds this support to the build matrix, enabling continuous integration of the next major Python version before its official release. This mirrors the earlier PR #157559 that enabled Python 3.14, showing PyTorch’s proactive approach to compatibility.

Free-threaded Python (3.15t) is particularly notable—it removes the global interpreter lock (GIL), allowing true parallel execution for CPU-bound tasks. By building nightly wheels for both standard and free-threaded Python 3.15, PyTorch lets ML engineers and framework maintainers validate performance and stability ahead of time. The manywheel format ensures broad Linux distribution compatibility. This early testing is critical for production environments that rely on PyTorch for inference and training workloads, as Python version upgrades often introduce breaking changes in C extensions and binary dependencies.