2026 Birthday week: network performance update

2026 Birthday week: network performance update | Cloudflare Blog

Skip to content

Cloudflare is now the fastest provider in 74% of the 1,000 largest networks around the world, up from 60%  in April 2026. This huge improvement matters because every millisecond affects how quickly users can reach the applications, APIs, and websites they rely on. In this Birthday Week performance update, we’ll review how we get our measurements, introduce a new measurement methodology using Cloudflare Challenge Pages, and discuss where these improvements have had the biggest impact for customers. Cloudflare is fastest in 74% of top networks In August, Cloudflare was the fastest provider in 74% of top networks, up 14 percentage points from our last update during Agents Week in April. The figure below shows the countries where Cloudflare is the fastest provider.

We improved from 60% to 74% by becoming the fastest provider in an additional 150 networks out of that top 1,000, and there are 38 additional countries where Cloudflare now ranks as the fastest. We measure this by looking at the fastest provider for users on the networks serving the largest number of users in each country. The graphic below shows countries where Cloudflare has become the fastest provider across those networks since April.

Here you see the number of additional networks on which Cloudflare is now the fastest.

How do we get these measurements? Our analysis begins with the 1,000 largest networks in the world, ranked by estimated user population using data from APNIC . Because these networks cover users across a wide range of geographies and access environments, they give us a useful view into how people actually experience the Internet. For each network, we evaluate performance using connection time: the amount of time required for a user’s device to complete a TCP handshake when requesting content. We use this because it maps closely to what people think of as a “fast” user experience. It reflects real-world factors such as distance, routing, and congestion, while still being specific enough for us to compare providers and identify where performance can improve. To rank providers, we calculate the trimean of connection times. The trimean combines the 25th percentile, 50th percentile, and 75th percentile into a weighted average. Using this method helps reduce the influence of unusual outliers while still representing the range of experiences that most users see. You can read previous posts  to learn more about why we chose this metric. When someone reaches a Cloudflare-branded error page, their browser can run a small background measurement that fetches lightweight files from several providers, including Cloudflare, Amazon CloudFront, Google, Fastly, and Akamai. We then record how long each connection takes from that user’s browser, on that user’s network, at that moment. This gives us a picture of performance under real Internet conditions, not just in controlled test environments. Since 2021, Cloudflare-branded error pages have provided a reliable source of real-user performance data, and they remain an important part of how we measure performance. But measurement gets better with scale. The more data we collect, and the more networks we observe, the more accurately we can understand how users experience the Internet. Since our last update, we have expanded our performance data collection by adding measurements using Cloudflare Challenge Pages. Introducing performance benchmarking with Challenge Pages If you're not already familiar, Cloudflare Challenge Pages are full-page screens that verify visitors before they reach a website. When a challenge is actioned — typically by a Web Application Firewall rule — the Challenge Page acts as a gate: it holds the request, evaluates the browser environment for automated signals, and only lets legitimate visitors through, usually with no interaction required. Challenge Pages are delivered by Cloudflare Turnstile , our privacy-preserving, risk-based challenge technology that runs directly in the visitor's browser. Because Challenge Pages run across a broad set of websites and real-world network conditions, they present a novel way to collect performance measurements from the places where people actually use the Internet. If you want to add Challenge Pages to your own website, you can get started with the Cloudflare Challenge Pages documentation . How does it work? From an end user's perspective, Challenge Page-based measurement works much like our existing error-page measurements: it runs quietly in the browser and does not require the user to do anything extra. While the visitor is on a Challenge Page a small, non-interactive measurement runs in the background. It fetches lightweight files from a fixed set of endpoints, including Cloudflare, Amazon CloudFront, Google, Fastly, and Akamai, and records whether each request completed and how long it took. While we care about measuring performance, we care even more about improving it. That includes making sure our measurements do not negatively affect the user experience. We designed the Challenge Page measurement to avoid adding noticeable latency for visitors, while preserving the privacy-focused properties that Turnstile brings to Challenge Pages and that make them different from traditional CAPTCHAs. For now, we are running measurements on only a small fraction of eligible free Challenge Pages, in addition to continuing to collect data from Cloudflare-branded error pages. We will limit these measurements to free Challenge Pages and we will only increase the sampling rate if the additional data improves measurement quality and end-user performance remains unaffected. Challenge Pages’ reach helps measurements scale The main upgrade from the error-page method alone is reach. Cloudflare-branded error pages have given us high-quality measurements from a narrower set of use cases, while Challenge Pages let us collect similar measurements during everyday interactions, wherever a challenge is already being served. That means more measurements from more networks, without requiring users to take any additional action. By collecting measurements on Challenge Pages, we are dramatically increasing the volume of data we collect and improving the diversity of users, networks, and geographies we measure. From a data quality perspective, this takes an already informative dataset to the next level. More measurement, more possibilities Even though the...