Cloudflare announced that its performance became the fastest across 74% of the 1,000 largest networks worldwide during August 2026, up from 60% in April. The company says the improvement reflects its leadership across 150 additional networks, along with progress in 38 additional countries, according to a methodology that compares users’ connection times with those of other infrastructure service providers.
The figure does not represent the launch of a new service so much as an update to a global performance metric and the way its data is collected. Cloudflare relies on APNIC estimates to rank the largest networks by number of users, then measures the time required to complete the TCP handshake when content is requested. The “Trimean,” which combines the 25th, 50th, and 75th percentiles with a specified weighting, is used to limit the impact of outliers and represent a broader range of user experiences.
How have the measurements expanded?
Since 2021, Cloudflare has collected part of its data from branded error pages. When a user reaches one of these pages, the browser can perform a small background measurement that fetches lightweight files from fixed points operated by Cloudflare, Amazon CloudFront, Google, Fastly, and Akamai, then records the time of each connection from the user’s network at the same moment.
The company has now added measurements from Cloudflare Challenge Pages, verification pages that appear before access to some websites when web application firewall rules are enabled. While the visitor remains on the page, a non-interactive background measurement runs without asking the user to take an additional step. Cloudflare says the design is intended to avoid adding noticeable time to the browsing experience while preserving the privacy characteristics associated with Turnstile, the risk-assessment-based verification technology.
What changes in practice?
Challenge Pages give the company access to broader samples of networks, countries, and everyday use cases, rather than relying on a narrower set of error pages. According to Cloudflare, this helps reduce the sensitivity of rankings to daily changes and outliers, particularly when only one or two milliseconds separate competing providers; the difference might be, for example, 50 milliseconds versus 51 milliseconds.
The larger volume of data also enables future analyses that go beyond counting networks, such as weighting performance by the number of people affected, aggregating results by country or globally, or estimating the share of user traffic covered by the measured networks. However, Cloudflare acknowledges that the initial results associated with Challenge Pages are still at an early stage, and that increasing the sample size will better determine whether a provider’s progress is genuine or the result of statistical noise.
Methodological limitations
The measurements are currently conducted on only a small percentage of eligible free Challenge Pages, in parallel with error-page data. The company says it will not expand the sampling rate unless doing so improves measurement quality without affecting user performance. Therefore, the 74% figure should be read as the result of Cloudflare’s methodology and the set of networks it selected, not as a comprehensive ranking of all internet traffic or an independent judgment covering every use case for content delivery services.
certi.news analysis: The most important value in the update is not the announced percentage alone, but the attempt to expand measurement from tests associated with error pages to broader interaction points across the web. This may make comparisons more stable, but it also ties the quality of the result to how Challenge Pages are selected, the sample size, and the potential impact of measurement on the experience. The comparison remains useful for infrastructure teams monitoring connection time, while requiring results to be examined by network and geographic location rather than relying solely on the global ranking.