Cloudflare opens BEACON, a BigQuery dataset of billions of real-user web performance measurements

What Cloudflare released

Cloudflare has released BEACON — Browser Experience Across Cloudflare's Observed Network — an anonymized dataset of billions of real-world performance measurements taken across 10,000 of the largest websites on its network. According to the Cloudflare blog, the dataset covers every major browser engine, is updated daily in Google BigQuery, and follows the schema defined by the community-led RUM Archive, a public Real User Monitoring database whose footprint Cloudflare says it expands roughly a hundred-fold.

The motivation Cloudflare gives is a perception gap: developers usually test on powerful laptops and flagship phones over fast Wi-Fi, while many end users are on four-year-old budget devices, throttled data plans or under-invested networks. BEACON is meant to replace that distorted view with objective data about how pages load for real people.

The dataset reports all three Core Web Vitals: Largest Contentful Paint (LCP) for load time, Cumulative Layout Shift (CLS) for visual stability, and Interaction to Next Paint (INP) for responsiveness. Instead of publishing a single P75 figure, Cloudflare publishes full histograms, so you can derive any percentile and examine the long tail where the worst experiences hide.

What the early numbers show

At P90 on iOS and Android, WebKit — currently the only browser engine on iOS — performs best overall, but the advantage is not universal. Cloudflare reports that in 46 countries where WebKit represents more than 10% of traffic, its LCP, INP, or both are at least 10% worse than Blink-based browsers such as Chrome, Edge and Opera. In Cambodia, WebKit accounts for 17.5% of page views, yet its LCP is 50% worse than Blink's.

The dataset also includes industry classifications from Cloudflare's Intel API. Government and Politics, Health, and Safe for Kids stand out as high-performing categories, while Ads, Religion, and Weather sites typically perform worst. The additional percentiles expose degradation that a single P75 result conceals, particularly for visual stability as measured by CLS.

Where loading time actually goes

BEACON extends the RUM Archive standard with LCP and INP sub-parts that break loading and interaction into distinct stages. The global LCP breakdown challenges a common assumption: downloading the LCP resource itself — an image, video or webfont — typically contributes the least to perceived loading time. In the Poor bucket, document TTFB averages 1891ms, load delay 1485ms, render delay 2002ms, but load duration only 119ms. For most page views that miss the Good threshold, the bigger opportunities are discovering the LCP candidate and unblocking its render. Cloudflare notes customers can address some resource-discovery delays with Smart Hints.

For INP, the breakdown separates input delay, processing time and presentation delay. In the Poor bucket these measure 84ms, 284ms and 217ms respectively: JavaScript processing dominates the slowest interactions, but presentation delay also rises sharply, which Cloudflare attributes largely to complex CSS layout recalculations. Customers can use tools such as Zaraz to reduce the impact of third-party JavaScript.

Single-page apps measured properly

Cloudflare has also added support for Chrome's Soft Navigations API, which provides accurate LCP reporting for client-side navigations common in single-page applications built with React, Vue, Angular or Svelte. The numbers show soft navigations rendering two to three times faster than hard navigations at every percentile: 274ms versus 791ms at P50, and 1816ms versus 4122ms at P95.

The trade-off is the landing page, which is often considerably heavier: 1370ms at P50, 2681ms at P75 and 8940ms at P95. Cloudflare's conclusion is that faster subsequent navigations must offset a slower first experience, and if users rarely progress beyond the landing page, the heavier initial load may never pay for itself.

Why it matters

Because BEACON is an open dataset, its value is not limited to the fields it ships with. Researchers can join it with external sources — Cloudflare's own example combines it with World Bank GDP per capita data to correlate economic conditions with web performance by country. Cloudflare Radar will also add a Web Performance section pairing BEACON with its Internet Quality Index, splitting the user experience into its two dominant components: the website being visited and the eyeball network delivering it. For developers, the immediate practical value is calibration. Rather than benchmarking against lab tests on fast hardware, teams can compare their Core Web Vitals against billions of real-user measurements and see precisely which stage of loading or interaction deserves optimisation — and, just as often, which one does not.