A speed test that ends inside your provider's network measures only the first leg: the shortest, cleanest stretch. Everyday browsing continues past it, across peering and transit, to servers on the open internet. Real-World Speed measures that longer, independent path, the same kind of route much of today's content travels, because the internet doesn't stop at your ISP. That is why it tracks what pages, videos and downloads actually feel like.
Neither ruler is universally right; they answer different questions. One estimates your connection's peak capacity. Ours holds one sustained connection across the open internet, closer to how everyday use feels.
On-net vs off-net
Most speed tests pick the nearest server they can reach, often one hosted inside your own provider's network, a few clean hops away. SpeedOf.Me connects to neutral CDN edge servers on the open internet, reached through the same interconnection points everyday traffic crosses; much of today's web is delivered exactly this way. Engineers call the first on-net and the second off-net.
Many streams vs one
Opening many parallel connections estimates the maximum capacity of your line: a legitimate, useful number. A single stream measures what one sustained transfer achieves, without stacking connections in parallel. We report the second, because it sits closer to the speed you feel.
All devices vs fixed-only
Ookla publishes separate fixed and mobile rankings; this comparison uses its fixed-broadband index. Atlas medians blend every test that runs on SpeedOf.Me: desktop and mobile, ethernet and WiFi, new hardware and old. That mix reflects how our testers actually test, and it reads lower than a fixed-line-only view.
Put the rulers side by side and the famous disagreements explain themselves. Ookla's fixed-broadband index measures peak performance to a nearby server; the Atlas measures everyday performance across the open internet. Neither column is wrong. They answer different questions.
| Country | Ookla rank | Ookla median | Atlas rank | Atlas median | Likely reasons |
|---|---|---|---|---|---|
| 🇸🇬Singapore | #1 | 425.46 Mbps | #20 | 63.0 Mbps | Multi-gig fiber plus nearby, often ISP-hosted test servers reward a parallel-stream sprint. One browser stream across the open internet lands mid-pack. |
| 🇦🇪United Arab Emirates | #2 | 384.51 Mbps | #45 | 44.4 Mbps | Multi-gig plans measured on-net. Off-net, a single stream travels much farther to everyday content. |
| ðŸ‡ðŸ‡°Hong Kong | #3 | 352.40 Mbps | #31 | 54.2 Mbps | World-class local fiber. Open-internet routes are long from here, and blended devices pull the median down. |
| 🇫🇷France | #4 | 352.35 Mbps | #10 | 90.1 Mbps | Strong on both rulers. |
| 🇮🇸Iceland | #5 | 347.30 Mbps | #7 | 97.5 Mbps | Strong on both rulers. |
| 🇨🇱Chile | #6 | 334.99 Mbps | #48 | 43.6 Mbps | A famous fiber build-out tops the on-net sprint; long transit legs to global content show up off-net. |
| 🇲🇴Macau | #7 | 318.05 Mbps | below our sample threshold | ||
| 🇺🇸United States | #8 | 309.80 Mbps | #1 | 143.5 Mbps | CDN edges and peering sit close to most users. Strong on both rulers; first on ours. |
| 🇨ðŸ‡Switzerland | #9 | 292.56 Mbps | #8 | 93.1 Mbps | |
| 🇩🇰Denmark | #10 | 291.60 Mbps | #4 | 125.5 Mbps | Strong on both rulers. |
Ookla figures: Speedtest Global Index, fixed broadband median download, March 2026, speedtest.net/global-index (accessed 2026-07-23). Ookla and Speedtest are trademarks of Ookla, LLC. Atlas figures: median real-world download, First half of 2026. The two sources cover different periods and methods; the notes are likely contributors, not causal findings from matched tests.
It is a panel, not a census. People test when they choose to, often when something feels slow. Medians over millions of tests are robust, but they describe the people who test, not every household in a country.
Geography leaks in. Every test runs against the best reachable server. Where the nearest edge is far away, or international routes are constrained, a country reads lower partly because of the route rather than its last mile. That is real experience, but it is worth naming.
One stream has a ceiling. On multi-gig plans, a single browser stream cannot show everything the line could carry in parallel. The very top of the market is undersold on our ruler, which is one reason multi-gig countries sit lower here than on on-net indexes.
Every speed index has a lens. We would rather show you ours than pretend there is none. The full method notes are in the H1 2026 report.
The track and the road
A closed test track is perfect: no traffic, no junctions, no weather. The road is where you actually drive. On-net speed tests are the track day. Real-World Speed is the road.
Shipped, delivered
A tracking page says your package left the warehouse within minutes. What you care about is when it reaches your door. Tests inside your ISP measure the dispatch. Real-World Speed measures the delivery.
Temperature, feels-like
The thermometer says 30 degrees; the forecast says it feels like 36. One reports the condition, the other reports the experience. A plan's headline Mbps is the condition. Real-World Speed is what it feels like to use.
Why is the United States #1 on this ranking?
Because of what we measure. Real-World Speed is a single browser stream to neutral servers on the open internet, blended across every device that tests. The United States leads at 143.5 Mbps because the everyday route from its users to online content is short and fast at the median: content networks and peering sit close to most people. Indexes that measure multi-stream sprints to servers inside each ISP reward raw plan capacity instead, and rank other countries first. Both are consistent; they answer different questions.
Why isn't Singapore #1 here?
Singapore sells some of the world's fastest fiber plans and tops on-net, multi-stream indexes like Ookla's. Measured as a single stream in the browser to neutral servers on the open internet, its median lands at 63 Mbps, #20 of 119 here. The plan capacity is world-class; what our ruler scores is the everyday route to open-internet content, on every device people actually test with.
Which number is right, ours or Ookla's?
Both, for different questions. A multi-stream test to a nearby server estimates your connection's maximum capacity: useful for checking you get the plan you pay for. Real-World Speed measures what a single sustained connection typically gets from the open internet: useful for predicting how browsing, streaming and downloads feel day to day. Peak performance and everyday performance are both real; they are just different numbers.
See your own Real-World Speed
Free, browser-based, no app. The same measurement engine behind every number in the Atlas, and the speed test API developers embed.
The country table is free to reuse under CC BY 4.0: CSV · JSON. The editorial story is in the H1 2026 report.