A fast connection should do more than produce a big download number. Our internet speed test helps you see how your connection sends data, receives data and responds when it is busy, so you can make a useful comparison rather than chase one impressive reading.
Give this test a quiet minute. It starts only when you ask, and you can stop it at any time. Keep the same network throughout the run.
Choose Quick for a lighter check, Balanced for everyday comparisons or Deep for a longer look at your connection. Their maximum total test data is 64, 240 and 600 MB respectively, plus connection overhead. These are caps, not amounts every test must use.
Close large downloads, pause backups and keep this tab active. Our test checks idle latency first, then download and upload. The graph records individual completed samples, including dips.
Read speed alongside latency under load. Export a branded PNG or raw CSV, and optionally save up to 5 completed tests in this browser. Retest after changing one thing, such as Wi-Fi distance.
Compare idle latency with upload-loaded latency. A sharp increase can explain why speech arrives late even when the headline speed looks healthy. Pause the backup, repeat the same depth and see whether responsiveness improves.
Run a Balanced test near your router, then repeat in the room where you work. Keep the device, network and test depth unchanged. If a wired comparison is strong but the distant Wi-Fi result falls, investigate the wireless link before buying a faster plan.
Start with Quick and read its 64 MB payload budget before running. Keep the phone still and avoid switching between Wi-Fi and mobile data mid-test. A phone with weak signal or a busy cell may behave differently a few streets away.
Enable local history and repeat at comparable times over several days. Export CSV for a spreadsheet or PNG for an easy-to-read report. Record your own connection type separately; our test does not pretend to identify your plan or certify an ISP service level.
Quick is useful when time or mobile data is limited. Balanced gives your connection more time to settle, while Deep takes a longer look at changing speeds. You can see the maximum data allowance before starting. The test may finish sooner if a data or safety limit is reached, and we flag a shortened result so you know how to interpret it.
Choose Test depth and Appearance beside Start test. Our analog dial follows the latest usable reading while a transfer is running, with a soft glow, a moving activity highlight and separate download and upload colors. The outer ring follows test progress. At the end, the needle settles on the final download result. Digital view keeps both speeds prominent. Switching views changes the presentation, not the measurement, and reduced-motion preferences are respected.
The graph helps you see whether your connection stays steady or varies during the test. The live dial can move as each transfer finishes; the final result summarizes the usable measurements rather than a brief peak. Read the speeds alongside idle and loaded latency for a more useful picture than one large number alone.
Idle latency tells you how quickly the connection responds before it is busy. Loaded latency shows what happens during downloads and uploads, while jitter describes how consistently those responses arrive. Lower, steadier readings are generally more useful for calls and games. A missing reading is unavailable, not zero.
Saving is off by default. When enabled, this browser keeps the latest 5 completed results; stopping a test does not create a final record. PNG includes the metrics, sample chart, date, depth and ClockTools link. CSV keeps numeric units in its headers for analysis. Exported files may reveal your network performance when you share them.
Our streaming, video-call and gaming indicators help you put the measurements in context. They consider the speeds and response-time readings relevant to each activity. They are broad starting points, not a service guarantee or a substitute for trying your own apps. Missing readings cannot establish that an activity will work well.
The Internet grew around packet switching: messages are split into pieces that share network links with other traffic. The Internet Society traces the early research, ARPANET and the transition to interconnected networks. The important idea for a speed test is sharing. Your data does not have a private road from your browser to every destination.
CERN dates the invention of the World Wide Web to 1989. The Web made linked documents accessible over the Internet; it did not invent the underlying network. Today that same network carries video, backups, remote desktops and games. Those activities make different demands, which is why one speed number cannot describe the entire experience.
A bit is a binary digit; a byte contains eight bits. Internet plans usually advertise megabits per second, while file managers often display megabytes per second. Divide a decimal Mbps figure by eight to get a theoretical MB/s equivalent: 80 Mbps becomes 10 MB/s.
For a fixed example, a 100 MB file at a perfectly steady 80 Mbps takes about 10 seconds before overhead. Real downloads also depend on the source server, protocol overhead, retransmissions, storage and competing traffic. Our completed-payload counter uses decimal MB; it is not a complete billable-data counter and does not include every network header.
80 Mbps ÷ 8 = 10 MB/s · 100 MB ÷ 10 MB/s ≈ 10 s
Throughput describes how much data moves over time. Latency describes waiting for a response. A broad motorway can carry many cars while one car still takes time to reach its destination. When uploads fill a router or access-network queue, small interactive requests may wait behind larger transfers.
That is why we probe during transfers, not only before them. A connection that responds quickly at rest but slows dramatically under load may struggle during calls or games while someone else backs up files. A larger loaded-latency value points to a symptom, not a diagnosis: routing, Wi-Fi contention, device load and the test endpoint can all contribute.
Test on the device and in the room where you actually use the connection. Then change one factor at a time: move nearer the router, use Ethernet, pause a backup or compare with and without a VPN if that is appropriate for your situation. Repeated observations are more informative than the best result from unrelated conditions.
Different speed tests can disagree without either being broken. Their endpoints, sample sizes, parallel connections and calculations may differ. Our test measures a browser-to-Cloudflare HTTP path with bounded sequential transfers; it does not saturate every connection with unlimited parallel downloads, predict every video service or measure the route to your game server.
ClockTools / Cloudflare. Our test checks your connection with real downloads, uploads and response-time measurements between your browser and a Cloudflare-hosted test endpoint. We look at repeated readings rather than treating a brief speed spike as the whole story. Latency and jitter add context, including how responsive the connection remains while it is busy. The result describes this device, connection and test route at that moment; it is not a promise of the same speed to every website. We show the data allowance before you start and identify results that need more measurement time.
A browser speed test is a snapshot, not a certified check of your internet plan. Wi-Fi, VPNs, other downloads and the route to the test server can affect it. It does not measure packet loss or guarantee how every app will perform. Test traffic uses generated data, not your personal files. The data total covers completed test transfers; unfinished requests and connection overhead can use additional data.
About your results · Correction policyNo. Transfers begin only after Start test. Changing the depth or view does not consume the test payload. Stop cancels the active measurement; a cancelled run is not saved as a completed result.
No. Test depth changes how long we check your connection and how much test data may be used, not the server location. You can see the actual test endpoint with your result and in exports. Speeds can differ across routes, so compare repeated tests under similar conditions.
Sometimes the connection, browser or available data allowance does not provide enough usable measurements for a final speed. We show -- rather than invent a number, and mark a shortened test when needed. Try again with a deeper test if your data allowance permits. An unsuccessful request is reported as an error, not a speed result.
The engine qualifies load from transfers lasting at least 250 ms. Very fast or brief transfers may not supply enough concurrent probes. Unavailable does not mean zero latency or a perfect connection.
Netflix currently recommends a stable 15 Mbps or more for one 4K stream. Other services, several simultaneous streams and background traffic can need more. Our streaming indication is only a broad starting point, not a promise that every video will play without interruption.
It measures the whole path from this browser to our endpoint, including your local connection. Compare Wi-Fi with Ethernet where possible to separate a local wireless bottleneck from a wider connection problem.
Optional history stays in this browser on this device, up to 5 completed tests. It is not an account or a cross-device archive. Clearing browser storage removes it. The test engine’s additional results-logging service is disabled; the site’s general analytics settings are unchanged.
No. The PNG and CSV are records of a browser test at a particular time. They can support a discussion with your provider, but they are not calibrated certification, continuous monitoring or proof of the performance of every destination.
For a measurement problem, tell us the page language, browser version, test depth, approximate time, connection type and which metric looked wrong. Include a screenshot or export if useful, but never send account passwords, private addresses or other secrets. We compare reproducible runs and check timing, payload and calculation errors. Verified corrections update the tool and its explanations; we cannot reconstruct a test you did not save or guarantee a fix to your ISP.
Report a measurement issue