Skip to main content

Internet speed test

Test your download speed, upload speed, and latency.

Free & unlimited
020406080100
0
Mbps
Ready
Processed on our servers and deleted right after. Never stored or shared.

This speed test runs three phases against an endpoint on this site, served from the Cloudflare edge location nearest to you. Latency comes first: ten sequential requests, each timed with performance.now, averaged into a ping figure with jitter calculated as the mean difference between consecutive samples. Download then opens six parallel streams requesting 25 MiB each, discards the first 1.5 seconds while the connection ramps up, and measures throughput over the following 8 seconds. Upload does the same with three streams of a 20 MiB block and a 2 second ramp. Speeds are reported in decimal megabits per second, the same unit ISPs advertise.

Key facts about Internet speed test

Key facts about Internet speed test
Test endpointthis site's own /api/speedtest routes, running as Cloudflare Pages Functions
Server locationwhichever Cloudflare edge location is closest to you — not a server you select
Ping methodten sequential no-store requests, each timed round trip with performance.now
Jittermean absolute difference between consecutive ping samples
Download streamssix parallel XHR streams, each requesting 25 MiB and looping until the window closes
Download window1.5 second ramp-up discarded, then 8 seconds of measurement
Upload streamsthree parallel POST streams of a 20 MiB block generated in your browser
Upload window2 second ramp-up discarded, then 8 seconds of measurement
Unitmegabits per second, decimal — bytes times eight, divided by one million
Download payloada fixed 64 KB filler buffer re-sent by the edge, so no CPU is spent generating random data
Upload payloadan empty 20 MiB ArrayBuffer created in the page — never one of your files
Capability panelchecks the result against HD 5, 4K 25, calls 3, gaming 15 plus ping under 50, backup 10, music 0.5, browsing 2, large downloads 50 Mbps

What happens to your file

This is the one tool on this page's family that does use the network, because measuring a network requires one. The requests go to this site's own /api/speedtest endpoints, handled by a Cloudflare Pages Function at the edge. What reaches that edge is what reaches any website you visit: the HTTP requests themselves, with your IP address, user agent and standard headers. The upload phase additionally sends a block of filler bytes created in your browser — an empty 20 MiB buffer, never a file of yours. The function counts the bytes it receives and discards them; nothing is written to a database, no result is stored, and there is no account, no history and no shareable server-side record. The result panel is computed in your browser, and the share button copies the numbers to your clipboard rather than posting them anywhere.

About this tool

  1. 1

    Stop other traffic first

    Pause downloads, streaming, cloud sync and video calls on every device on the network. The test measures what is left of the line, so anything else running counts against your result.

  2. 2

    Choose wired or wireless deliberately

    Ethernet measures the line itself. Wi-Fi measures the line and the wireless link together, and on a busy 2.4 GHz channel the wireless link is usually the limit.

  3. 3

    Start the test

    Ping runs first, then download, then upload. Each phase shows a live gauge and a rolling chart, so you can see whether throughput is steady or spiky.

  4. 4

    Ignore the first seconds

    The tool already discards the ramp-up window before measuring, which is why the gauge moves before any number is recorded. TCP takes a moment to reach full speed on every connection.

  5. 5

    Read ping and jitter together

    Ping is the average round trip to the nearest Cloudflare edge. Jitter is how much consecutive samples varied, and it is the figure that predicts whether calls and games will feel stable.

  6. 6

    Compare against your plan

    ISPs advertise in megabits per second and so does this test, so the numbers are directly comparable. Expect to land somewhat below the headline figure; protocol overhead alone costs a few per cent.

  7. 7

    Repeat before concluding

    Run the test three times, a few minutes apart, and use the middle result. A single measurement catches whatever momentary congestion happened to exist.

Specs & compatibility
Browser supportany browser with XMLHttpRequest progress events — all current desktop and mobile browsers
Per-request capsthe edge function serves at most 100 MB per download request and accepts at most 100 MB per upload
Total transferredtypically 100 to 800 MB per run on a fast line — watch out on a metered connection
Test durationroughly 22 seconds: ping, then 9.5 seconds download, then 10 seconds upload
Practical ceilingvery fast lines can be limited by the browser and the six-stream setup rather than by the connection
Mobile browserswork, but background app refresh and radio power states add variance
Corporate networksproxies and inspection appliances sit in the path and will cap the measurement
Account or installnone — no signup, no app
  • A run can move several hundred megabytes. On a mobile data plan or a metered connection, check your allowance before testing repeatedly.
  • Wi-Fi 5 and older access points often cap real throughput near 300 to 400 Mbps regardless of the line behind them; test over Ethernet before blaming the ISP.
  • Jitter under about 5 ms is comfortable for video calls and gaming. Consistently high jitter with a normal ping usually points at Wi-Fi interference or a congested upstream.
  • Upload is deliberately measured with fewer streams than download because most consumer lines are asymmetric and saturate on fewer connections.
  • A laptop on battery saver can throttle the network stack; plug it in for a clean measurement.
  • VPNs, corporate proxies and security appliances all terminate and re-open the connection, and every one of them lowers the ceiling.
  • If ping is good but download is poor, suspect Wi-Fi or a device limit. If ping is poor too, the problem is more likely on the line itself.
  • Test at 8 pm as well as at midday when documenting a complaint to an ISP — evening congestion is exactly what they will ask about.
  • Ping and jitter from ten timed round trips
  • Six-stream download measurement with a discarded ramp-up window
  • Three-stream upload measurement with its own ramp-up window
  • Live analogue gauge and rolling throughput chart
  • Results in megabits per second, matching ISP advertising
  • Capability panel for streaming, calls, gaming and backup
  • One-click copy of the result text
  • Checking whether your connection delivers the speed your ISP plan promises.
  • Working out whether a slow video call is caused by upload bandwidth or by jitter.
  • Comparing throughput before and after changing a router, a cable or a Wi-Fi band.
  • Testing each room in a house to find where the wireless signal falls off.
  • Collecting evidence at different times of day before raising a support ticket.
  • Confirming that a new fibre or cable installation actually performs as sold.
Only what any web request sends, plus filler. The ping and download phases are ordinary GET requests; the upload phase posts an empty 20 MiB buffer created in your browser, not a file from your computer. The Cloudflare Pages Function at the other end counts the bytes and throws them away. Your IP address and standard headers reach that edge exactly as they do for every page you load. No result is written to a database and there is no account, history or shareable link stored anywhere.
This site's own endpoint, served by Cloudflare's edge network, which routes you to whichever of its locations is nearest. There is no server picker, because there is only one logical endpoint and Cloudflare decides which location answers. That keeps latency honest for measuring your own connection, but it means you cannot use this test to measure the route to a specific distant city, which is something a multi-server tool is better suited to.
A gap of five to ten per cent is normal: TCP, TLS and HTTP overhead all consume bandwidth that never shows up as payload. Larger gaps usually come from Wi-Fi rather than the line, from another device on the network, from an older router, or from the device's own network adapter. Test over Ethernet with everything else paused to separate the line from the local network. If a wired test still falls well short at several times of day, that is the evidence worth taking to the ISP.
Ping is the average time a packet takes to reach the nearest edge and come back. Jitter is how much that time varies between consecutive samples. Real-time applications buffer against latency, so a steady 60 ms is perfectly usable for a call; what they cannot hide is variability, because packets arriving out of rhythm produce the choppy audio, frozen frames and rubber-banding you notice in games. A low ping with high jitter feels worse than a higher ping that is stable.
It depends on your speed, because the phases run for a fixed time rather than a fixed size. The download phase measures for 8 seconds after a 1.5 second ramp across six streams, and the upload phase for 8 seconds after a 2 second ramp across three. On a 100 Mbps line that is roughly 120 MB in each direction; on a gigabit line it can exceed half a gigabyte. On mobile data or any metered connection, run it sparingly.
Not reliably. Above roughly one gigabit the limiting factor is often the browser, the number of parallel streams and the device's own network stack rather than the line. Six streams from a single browser tab will usually saturate a gigabit connection but may not saturate 2.5 or 10 gigabit. If this test reports a figure well below a multi-gigabit plan, treat it as a floor rather than a verdict, and confirm with a native tool on a machine wired directly to the router.

Related tools

View all

Updated

We use anonymous analytics to improve ToolChamp. No personal data is stored or sold. Privacy Policy