# Kubernetes v1.37: Native Histograms Graduates to Beta

DevFeed: [Kubernetes v1.37: Native Histograms Graduates to Beta](<https://devfeed.tech/articles/kubernetes-v1-37-native-histograms-graduates-to-beta-4583.md>)

Original publisher: [Read original article](<https://kubernetes.io/blog/2026/09/11/kubernetes-v1-37-native-histograms-beta/>)

Author: Richa Banker

Published: 2026-09-11T18:30:00Z

Content type: release

Language: en

Sources: [Kubernetes Blog](<https://devfeed.tech/sources/kubernetes-blog.md>)

Topics: [telemetry](<https://devfeed.tech/topics/telemetry.md>)

Tags: [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [latency](<https://devfeed.tech/tags/latency.md>), [metrics](<https://devfeed.tech/tags/metrics.md>), [observability](<https://devfeed.tech/tags/observability.md>), [prometheus](<https://devfeed.tech/tags/prometheus.md>), [storage](<https://devfeed.tech/tags/storage.md>), [telemetry](<https://devfeed.tech/tags/telemetry.md>), [time-series](<https://devfeed.tech/tags/time-series.md>)

## AI overview

Kubernetes v1.37 enables Prometheus native histograms by default at Beta. The article explains how dynamic exponential buckets improve metric resolution and quantile accuracy while reducing time-series, scraping, and storage overhead compared with classic histograms.

## Source excerpt

I'm excited to announce that native histogram support for Kubernetes metrics is graduating to Beta and is enabled by default in Kubernetes v1.37! Native histograms (previously introduced as Alpha in Kubernetes v1.36 under KEP-5808) bring high-resolution, low-cardinality observability to Kubernetes metrics. By adopting Prometheus Native Histograms, Kubernetes components now expose latency and duration metrics with far greater accuracy while significantly reducing telemetry storage and scraping overhead. Why move beyond classic histograms? Since the early days of Kubernetes observability, duration and latency metrics (such as API server request latencies or scheduling durations) have relied on classic Prometheus histograms. Classic histograms require metric authors to define a static list of cumulative bucket boundaries (le labels), such as 0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10. While familiar, this approach introduces three major challenges: The Bucket Guessing Game: If a workload's latency profile changes, for example, shifting into microsecond ranges or experiencing long-tail tail latencies beyond the highest bucket, the histogram loses visibility. Specifying bucket boundaries upfront requires knowing the distribution before observing it High Cardinality & Storage Cost: With classic histograms, each bucket boundary is exported as a separate time series (_bucket{le="..."}). A histogram with 10 buckets across multiple labels multiplies the number of time series by 10, increasing memory consumption in Prometheus and inflating time series database (TSDB) storage costs Interpolation Error in Quantiles: Calculating percentiles using histogram_quantile() relies on linear interpolation between static bucket boundaries. When bucket spans are coarse, quantile calculations can suffer from significant estimation error What are Prometheus native histograms? Prometheus Native Histograms replace static user-defined buckets with dynamic, exponential buckets. Inst