API Response Time Percentile Calculator | P50 to P99 in One Paste
Paste a list of API or web server response times and get P50, P90, P95 and P99 percentiles alongside mean, median, standard deviation, min, max and a distribution histogram. It surfaces the tail latency that a plain average hides, right inside your browser.
💡 About this tool
If you describe your service speed as "averaging 80 ms," you quietly bury the slow requests a small slice of users actually feel. Nine fast responses out of ten mean nothing to the one user stuck waiting on the tenth, and that slow request is usually what triggers support tickets and SLA breaches.
A percentile tells you what share of your data falls at or below a value. A P95 of 320 ms means 95% of requests came back within 320 ms. Backend engineers and SREs lean on P95 and P99 instead of the average precisely because high percentiles track perceived experience far better than the mean does.
Drop in a column of response times exported from your logs and this tool reports the four headline percentiles plus mean, median, standard deviation, min, max and count. The histogram shows the shape of the distribution at a glance, so you can tell a tight cluster from a long, dragging tail.
🧐 Frequently Asked Questions
How should I format the input? One value per line, or separated by commas or spaces. Non-numeric characters and blank lines are ignored automatically, so pasting straight from a log column works fine.
Do the values have to be in milliseconds? The math runs on whatever numbers you paste, so seconds or microseconds work too. The display label is fixed to "ms," so keep your units consistent when you read the results.
How are the percentiles calculated? The tool sorts your data and uses linear interpolation between the nearest ranks. When a percentile position falls between two data points, it blends them proportionally rather than rounding to a single sample.
Why are my mean and median (P50) so far apart? In a long-tailed distribution, a handful of very large values drag the mean upward while the median barely moves. The wider the gap, the more skewed your data is toward slow outliers.
Why is the histogram not showing up? A distribution needs at least two data points to bucket. Paste several values together rather than a single number and the chart will render.
📚 Why P99 gets so much attention
In a high-traffic service, a single user often fires dozens of API calls in one session. Even if P99 only means "slow one time in a hundred," a screen that strings together twenty calls has a much higher chance of hitting at least one slow request every time it loads. That compounding is exactly why tail latency, not the average, shapes how fast a product feels.
That is why operations teams monitor high percentiles like P95 and P99 rather than the mean. Read alongside the standard deviation and histogram, percentiles let you separate "uniformly slow" from "usually fast but occasionally spiking," which is half the battle when you are hunting a bottleneck.