Python 3.12 performance improvements worth knowing

The JIT compiler is experimental but showing 4-5% speedups in tight loops, f-strings now support nested expressions and multi-line formatting without backslashes, a new type alias syntax type Point = tuple[float, float] replaces older TypeAlias for better readability.

The PEP 684 per-interpreter GIL is probably the most significant change for long-running services rather than pure execution speed — it lets you run multiple Python interpreters in a single process with independent GILs, which means true parallelism for IO/CPU workloads that were previously stuck waiting on one lock.

On raw performance:

• Tier 2 JIT (PEP 744) is still experimental but the infrastructure is there. For most code this won't matter yet, but it's a non-trivial foundation to have shipped in 3.12.
• Faster CPython += 10-60% on many workloads — mostly from internal micro-optimizations like specialized adaptive interpreter instructions and reduced object overhead.
• The f-string syntax PEP 701 lets you do expressions, comments, and multi-line formatting inside {} which is a quality of life win but also eliminates a lot of string concatenation boilerplate that the bytecode has

For anyone still running 3.10 or earlier, the f-string optimizations and PEP 709 (comprehensions inlined) make a noticeable difference on hot paths without changing any code. The JIT is experimental but worth tracking for data-heavy workloads — I've been testing it on an image processing script and saw roughly 5-10% improvement after the warmup period.

One thing to watch: some of the type hinting refinements in 3.12 can break older linters if you're using strict mode with a static analysis tool that hasn't caught up yet — upgraded my Ruff config last week before shipping and it was fine but worth checking your CI pipeline first.