Python 3.15.0 arrived on October 9, 2026, and the headline number is the JIT compiler. The experimental tier-two JIT now delivers a 7-8% geometric mean performance improvement on x86-64 Linux over the standard interpreter, and 11-12% on AArch64 macOS over the tail-calling interpreter. These are not benchmarketing numbers cherry-picked from a microbenchmark — geometric mean across the standard suite is the honest metric, and double-digit gains on ARM are meaningful for the growing share of developers on Apple Silicon. The official Windows 64-bit binaries now use the tail-calling interpreter, and macOS ships free-threading support by default. The performance story is no longer aspirational. The language-level additions are pragmatic rather than flashy. PEP 810 introduces explicit lazy imports, targeting the startup-time problem that has plagued Python CLI tools and serverless functions for years. PEP 814 adds frozendict as a built-in type — a request that's been floating in the community for over a decade. PEP 661 adds a sentinel type, and PEP 798 brings unpacking in comprehensions. PEP 829 introduces package startup configuration files, giving library authors control over import behavior without monkey-patching. These are infrastructure fixes, not syntax sugar. The typing system continues its steady march. PEP 728 adds TypedDict with typed extra items, PEP 747 introduces TypeForm for annotating type forms, and PEP 800 brings disjoint bases. These matter because they close gaps that forced workarounds in large codebases using mypy or pyright. The new profiling package (PEP 799) and Tachyon statistical sampling profiler are tools aimed at production Python — the kind of instrumentation that was previously third-party territory. The C API work deserves attention from anyone building extensions. PEP 782's PyBytesWriter API, PEP 788's finalization protection, and the stable ABI for free-threaded builds (PEPs 803, 820, 793) are the unsexy plumbing that determines whether the free-threading experiment actually gets adopted by the extension ecosystem. Frame pointers enabled by default (PEP 831) is a nod to systems-level observability — useful for perf, eBPF, and production profiling on Linux. The macOS 27.0 tkinter hang is a notable caveat. A Tk graphics toolkit incompatibility with the latest macOS means IDLE and other tkinter apps may freeze when opening dialogs. The advisory to defer macOS 27.0 upgrades if you depend on Tk-based apps is unusually blunt for a release note. Issue #158053 tracks the fix. The release is backed by 1,012 contributors and funding from the Sovereign Tech Agency, which supported Hugo van Kemenade's work as release manager through the Sovereign Tech Fellowship. The community celebration includes Barry Warsaw's 'whatsnewt' — a TUI text adventure game where every puzzle requires using an actual 3.15 feature in a live interpreter. It's a clever onboarding device and a sign of a project that still enjoys its own culture. Python 3.15 is an infrastructure release. The generative value is in the JIT maturation, lazy imports for startup performance, free-threading going mainstream on macOS, and the C API stabilization that will determine whether the free-threading story actually works at ecosystem scale. No single feature is revolutionary; the compound effect across performance, typing, and extension support is what matters.
