December 25, 2025
Is that py‑spy in a trench coat?
Tachyon: High frequency statistical sampling profiler
Python’s new Tachyon: built‑in speed tracker has devs yelling “py‑spy in the box”
TLDR: Python 3.15 adds Tachyon, a built‑in tool that samples running code to reveal slow spots—even on live apps. The big comment mood: “py‑spy in the stdlib,” mixing hype for easy, low‑overhead profiling with reminders it’s an estimator, not exact timing—useful, accessible performance sleuthing for everyone.
Python just dropped Tachyon in 3.15—a built‑in “statistical profiler,” aka a speed detective that peeks at your program many times per second to estimate where time really goes. It can even attach to a running app without restarts and claims “virtually zero” slowdown. The docs show sample‑based estimates (think: thousands of tiny snapshots) and caution it’s not a stopwatch—short scripts may need the older deterministic tracer. Source lives here.
But the internet’s headline is the comments: the top vibe is basically, “they put py‑spy in the box.” Fans are cheering the end of fiddly installs and the arrival of safer, production‑friendly profiling right in standard Python. Doubters are side‑eyeing the “near‑zero overhead” claim and reminding everyone that statistics wiggle—numbers will drift a bit from run to run. Meanwhile, memes are sprinting: “CSI: Python” jokes, “Find My CPU” quips, and speed‑of‑light puns about “Tachyon” not actually making code faster. The mini‑drama: does this overshadow existing tools or finally push more teams to profile in production? Either way, the mood is electric—Python just shipped a built‑in performance detective, and devs are ready to throw their slow scripts under the microscope.
Key Points
- •Tachyon (profiling.sampling) is a statistical sampling profiler added to Python in version 3.15.
- •It can run scripts or attach to running Python processes without code changes or restarts.
- •Sampling occurs externally to the target process, resulting in virtually zero overhead suitable for development and production.
- •Time is estimated from sample counts multiplied by the sampling interval; accuracy improves with more samples and longer durations.
- •For very short scripts (under one second), the deterministic profiler profiling.tracing is recommended instead.