Internet Speed Test

Measures download and upload throughput and latency by moving real data between your browser and Cloudflare’s nearest edge. Download and upload are separate measurements over the network — not one derived from the other.

—ready

—

Key takeaway: a browser speed test reads a little lower than a dedicated app, especially on very fast lines, because it is limited by what JavaScript and HTTPS can push through. Use it to compare your connection against itself — now versus later, Wi-Fi versus cable — rather than to hold your provider to the exact number on your contract.

What is actually being measured

Download opens several parallel connections to Cloudflare’s nearest edge and reads as fast as your line delivers. Parallel matters: a single connection often fails to fill a fast link because of how TCP ramps up, which is why a one-stream test can report half your real speed on fibre. The first second and a half is discarded so the ramp-up is not counted against you.

Upload does the mirror image — it POSTs randomly generated data to the same service and times how long it takes to leave. Random data matters, because compressible data would be squeezed in transit and produce a flattering, meaningless number.

Latency is measured with tiny requests, taking the median rather than the average so one stalled request does not distort it. Jitter is the average variation between consecutive measurements. If you care mainly about latency, the ping test measures it in more detail over a longer run.

Why the previous version of this page was replaced

Its upload test never touched the network. It opened two WebRTC connections inside the same browser tab and sent data from one to the other, which measures how quickly the browser can copy memory between two objects on your own machine. That number is roughly the same on gigabit fibre and on a throttled mobile connection. When the mechanism failed, the fallback reported 90% of the download figure as the upload speed, or a flat 10 Mbps — invented numbers, presented as measurements. It is fixed here: if a measurement cannot be taken, the page says so instead of estimating one.

Reading your result

Roughly	Comfortable for
Under 10 Mbps	Browsing and standard-definition video. HD streaming will struggle if more than one person is online.
25–50 Mbps	HD and 4K streaming, video calls, a household of several people.
100–300 Mbps	Large downloads and many simultaneous devices. Beyond this the bottleneck is rarely the line.
500 Mbps and up	Genuinely useful for moving very large files. For streaming and browsing, no different from 200.

Upload is the number people overlook and the one that matters for video calls, cloud backups and sending large files. Cable and most DSL connections are deliberately asymmetric — 500 Mbps down and 20 up is a normal shape — while fibre is often symmetric. If your calls break up while your downloads are fine, upload saturation is the usual reason.

Why your result might look wrong

Wi-Fi, almost always. The test measures the whole path, and the weakest link is usually the wireless hop. Testing on a cable is the fastest way to tell whether the problem is your line or your Wi-Fi — and if cable is fast while Wi-Fi is slow, nothing your provider does will help.

Something else is using the connection. A cloud backup, a game update or another person streaming will take their share.

The device itself. An older phone or laptop can be the limiting factor on a fast line, particularly over Wi-Fi.

A VPN is on. Encryption and the detour through another server both cost throughput. Test with it off for a baseline, then on, and compare.

If the numbers stay low on a cable with nothing else running, work through the wider diagnosis and check whether your provider is throttling before calling them.

Testing whether your provider throttles

Some providers slow specific traffic — streaming or large transfers — while leaving speed tests untouched, which is why a test can look fine while video buffers. A VPN hides what the traffic is, so if the same activity is noticeably faster through one, throttling is the likely explanation. That comparison is a genuine diagnostic, not just a reason to buy a VPN.

PureVPN How to detect throttling

Affiliate link. We do not rank by commission — compare providers yourself.

What this test cannot tell you

Your speed to anywhere else. It measures the path to Cloudflare’s nearest edge, which is deliberately close and fast. A slow game server or a distant download mirror is a different route entirely.

Whether your provider is meeting its contract. Advertised speeds are usually “up to” and measured to the router over a cable. A browser test over Wi-Fi is not the evidence a complaint needs — most regulators expect a wired test, repeated at different times.

Why a specific site is slow. That is usually the site, the route to it, or DNS — check the DNS checker and ping test instead.

Frequently asked questions

Why is this slower than my provider’s own speed test?

Provider tests usually run against a server inside their own network, so the traffic never crosses the wider internet. That is a legitimate measurement of your line, and also the most flattering possible one. A test to an outside network shows what you actually get when using the internet.

How much data does this use?

As much as it can move in the time allowed, which depends on your speed. The page reports the exact figure when it finishes. On a mobile connection, use the quick test — it caps the transfer far lower.

Why does the number jump around while it runs?

Because throughput genuinely varies, and early readings average in the connection ramp-up. The final figure is taken from the steady period after that, which is why it often differs from the peak you saw mid-test.

Should I test several times?

Yes, and at different times of day. Evening congestion is real, and a single measurement tells you about one moment. Three tests across a day are far more informative than one.

Does a VPN slow everything down?

It costs something — encryption plus a longer path. A nearby server on a good provider might cost you ten percent; one on another continent can halve your throughput. If the loss is dramatic, try a closer server before blaming the provider.