2026 Birthday week: network performance update

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 initial results from Challenge Pages measurements are promising, we have even more ideas for how to improve the dataset. First, we want our measurements to describe the experience of as many users as possible. Because Challenge Pages runs across such a broad set of websites and visitors, it gives us measurements from networks well beyond the top 1,000 and far more samples within each one, especially as we increase our test volume. That breadth gives us more analytical options: we can explore views that a network-count ranking alone cannot support, such as weighting performance by the number of people who experience it, grouping results by country or worldwide instead of by network, and quantifying how much of total user traffic our measured networks represent.  

Second, more measurements give us sharper resolution where the race is closest. In many networks, the top providers are separated by only a millisecond or two of trimean connection time such as Cloudflare at 50 ms and Fastly at 51 ms, which is a gap small enough that ordinary day-to-day variation can flip the ranking. As Challenge Pages add measurement volume, the confidence interval around each provider's trimean will narrow, letting us distinguish a genuine lead from statistical noise. The results we are sharing today are early, but as the dataset becomes larger, we expect future movement in these rankings to further reflect real changes in performance.

Improved measurements show Cloudflare as #1

This significant improvement coincides with the addition of our new measurement methodology described above. By increasing measurement volume, Cloudflare has more opportunities to compare performance against other top providers across a broader set of networks and user conditions. In practice, this gives us a clearer signal in networks where performance among top providers is very close.

For example, Cloudflare may have previously ranked second in some networks even though our trimean connection time was only 1 or 2 ms slower than the fastest provider. With more measurements, the results are less sensitive to outliers and day-to-day variation. That makes rankings more stable, especially in countries and networks where the fastest provider may have previously changed from one day to the next. As the dataset becomes larger and more representative, we get a more consistent view of which provider is actually fastest in each network.

Performance is a process

Improving performance is a continuous process, and so is improving how we measure it. This year’s results show meaningful progress: Cloudflare is now the fastest provider in 74% of the top networks we measure, and our new Challenge Pages-based methodology gives us a broader, more stable view of Internet performance around the world. We’ll keep using that data to find where we can be faster, validate the impact of our improvements, and make the Internet better for the customers and users who rely on Cloudflare every day.

Follow our blog for more performance updates as we continue to make the Internet faster.