Skip to content- Sources: primary, discussion
- Summary: Miguel Grinberg ran recursion-heavy and loop-heavy microbenchmarks across CPython 3.10 to 3.15, plus JIT and free-threading builds, PyPy, Node and Rust. The standard 3.15 interpreter is only marginally faster than 3.14 and slightly slower on one multi-threaded test, so there is no performance reason to rush an upgrade. The new result is the JIT, which for the first time beats the standard interpreter of the same version by 1.20x to 1.28x, having previously trailed it, while free-threading 3.15 runs about 4.5 times faster than the standard interpreter on the multi-threaded recursion test and still trails it on single-threaded work.
- Why it matters: The author labels the benchmark informal, describes the two programs and the averaging method, notes he tested 3.15.0rc3 rather than the final release, separates his upgrade opinion from the measurements, and recommends against the still-experimental JIT in production.