---
title: "Percentile"
description: "A percentile is the value below which a given percentage of a dataset falls — for example, the 95th percentile (p95) of response times is the value that 95% of requests were faster than."
canonical: "https://motherduck.com/glossary/percentile/"
related:
  - title: "DuckLake Lakehouse: Getting Started to Going Fast | MotherDuck"
    url: "https://motherduck.com/videos/ducklake-lakehouse-getting-started/"
  - title: "MD_LIVE_DUCKLING_SIZE | MotherDuck Docs"
    url: "https://motherduck.com/docs/sql-reference/motherduck-sql-reference/md-live-duckling-size/"
  - title: "PyPi Data | MotherDuck Docs"
    url: "https://motherduck.com/docs/getting-started/sample-data-queries/pypi/"
gated_asset:
  title: "DuckLake on MotherDuck"
  url: "https://motherduck.com/product/ducklake/"
---

# Percentile

> A percentile is the value below which a given percentage of a dataset falls — for example, the 95th percentile (p95) of response times is the value that 95% of requests were faster than.

## Overview

Percentiles describe the distribution of a dataset by identifying the value at a given rank. The 50th percentile (p50) is the median; the 95th percentile (p95) and 99th percentile (p99) are especially common in performance and reliability monitoring because they capture the experience of the slowest requests, which averages hide. A system with a fast average latency can still have a terrible p99 if a small fraction of requests are very slow — percentiles surface that in a way a mean cannot.

## Computing percentiles in DuckDB

DuckDB supports exact percentile calculation through `quantile_cont` (continuous, interpolated between values) and `quantile_disc` (discrete, returns an actual value from the dataset), along with a dedicated `median` function:

```sql
SELECT
    quantile_cont(latency_ms, 0.5)  AS p50,
    quantile_cont(latency_ms, 0.95) AS p95,
    quantile_cont(latency_ms, 0.99) AS p99
FROM requests;

SELECT median(latency_ms) FROM requests;  -- shorthand for the 0.5 quantile
```

`quantile_cont` and `quantile_disc` also accept a list of fractions to compute multiple percentiles from a single pass over the data:

```sql
SELECT quantile_cont(latency_ms, [0.5, 0.95, 0.99]) AS percentiles
FROM requests;
```

## Exact versus approximate

Exact percentile calculation requires sorting (or an equivalent ordering operation) over the full column, which can be expensive on very large datasets. When an estimate is acceptable — dashboards, monitoring, exploratory analysis — DuckDB's `approx_quantile` function (based on a t-digest sketch) computes percentiles much faster and with bounded memory, at the cost of a small, generally negligible amount of accuracy.