Tool review: September 10, 2026.
Paste latency samples in milliseconds to calculate common percentiles. The tool is useful for learning how a distribution behaves and for checking a small dataset, but production percentile systems may use streaming histograms or approximation methods that differ from a simple sorted sample.
What a percentile means
If p95 is 800 ms, 95% of the observations in the sample are at or below the reported p95 value and the remaining observations are slower. A percentile describes a position in a distribution; it does not mean that 95% of requests took exactly that long.
Why averages can mislead
Suppose most requests complete near 100 ms but a small group takes several seconds. The average moves upward, but it does not show how many users are in the slow tail. Percentiles expose that shape more directly and are therefore common in service-level indicators and performance dashboards.
How to compare percentiles correctly
- Use the same population and time window.
- Separate routes or operations with very different latency profiles.
- Keep failed requests visible instead of silently dropping them.
- Know whether the monitoring backend computes exact or approximate quantiles.
Small samples need caution
With only a handful of observations, high percentiles can be unstable because one value has a large influence. Use this calculator for education and small checks; rely on a production telemetry pipeline for representative traffic.
Related resources
Why the percentile needs a population
A percentile only has meaning relative to the observations included in it. Keep the time window, endpoint, region and client population stable when comparing p95 or p99 values. If traffic mix changes, investigate the distribution before concluding that the service itself became faster or slower.