Python 3.14 subinterpreters cut a CPU-bound job from 3.6 to 0.9 seconds, but reject concurrent calls

A developer benchmarked a CPU-bound prime-counting task in a Python 3.14 container with four cores. Four subinterpreters from the new concurrent.interpreters module finished in about 0.9 seconds, close to a single-thread baseline and far faster than four classic GIL-bound threads at 3.6 seconds. Subinterpreters performed comparably to multiprocessing in the same tests. However, a single interpreter rejects a second concurrent call with an InterpreterError rather than queuing it. The author concludes that running N tasks in parallel requires N separate interpreters, so thread-pool patterns do not carry over directly.