Python 3.15: JIT Gives 7–8%, but Lazy Imports and Profiler Matter More

Python 3.15.0 shipped on October 9, 2026. The JIT delivers a 7–8% geometric mean speedup on x86-64 Linux and 11–12% on AArch64 macOS, but numpy, pandas and Arrow code does not benefit. For data teams the real changes are PEP 810 lazy imports, the Tachyon sampling profiler and PEP 829 .start files.

• JIT: +7–8% on x86-64 Linux and +11–12% on AArch64 macOS, Python-level code only
• Pure-Python 1M-row loop about 8% faster; row-wise apply only 3–4% faster
• Vectorized groupby().sum() on 2M rows stayed within noise
• PEP 803 adds abi3t: Stable ABI extensions now work with free-threaded builds