Popular tools

Theme

Website Speed Test

How fast is your website? Test it from up to 3 places at once on a real browser, see what a visitor saw moment by moment, and get fixes for your own stack.

Lighthouse 13 · Chrome · 9 places · median of 3 runs · free, no sign-up
Every test is fresh: a new browser, an empty cache, nothing reused from an earlier test.9 places
What a visitor saw from the first byte to fully loaded
RequestsDocumentScriptCSSImageFontFetchMediaOtherLong tasksLayout shifts
The strip replays the loadafter a test
The screen every few hundred milliseconds, every request by type, the jobs that block taps and every layout shift, on one time line. Click any moment and it reads it to you: when it happened, what it was waiting for, and how long each part took.
Website speed test result: the grade, the six metrics and the strip of what a visitor saw while the page loadedWebsite speed test result: the grade, the six metrics and the strip of what a visitor saw while the page loadedAny pageA domain or a full URL; up to 3 places, phone, desktop or both.The gradeThe median run's Lighthouse score as a letter, with the runs on a 0–100 track.What a visitor sawThe screen, every request, long tasks and shifts on one time line: click any moment.The conditionsPlace, device, network, CPU, Chrome and Lighthouse versions on every report.
A result: the grade from the median of 3 runs, the six metrics with the points each cost, and the strip that replays the load.

How does a website speed test work?

In a page speed test, a test server in the place you pick opens the page in a fresh Chrome, three times, and records everything: the screen, every request, the main thread. Lighthouse turns each load into a score, and the median of the three runs is the grade.

Parts that need no browser arrive first: server timing from our 11 places, real users' data from the Chrome UX Report, and the site's stack. The grade follows when every place has finished its runs; the fixes come right after.

  1. 1SafetyCloudflare's malware and adult lists, before anything opens the page
  2. 2Server timingDNS, connect, TLS and first byte from 11 places, no browser
  3. 3Real users28 days of Chrome UX Report data for the page or its site
  4. 4Stackthe CMS, framework, host and CDN from our own engines
  5. 5Runs3 Lighthouse runs per place, each streamed as it finishes
  6. 6Gradethe median run's score as a letter, with the points each metric cost
  7. 7Fixeswritten for this stack, each file and number checked against the test
The parts of one test, in the order they reach the page.

What do LCP, CLS, TBT, FCP, Speed Index and TTFB mean?

Six numbers on one time line. TTFB is when the server's first byte arrived, FCP when the first text or image appeared, LCP when the largest thing on the first screen was drawn, and Speed Index how quickly the screen filled in.

TBT adds up how long the main thread was too busy to answer a tap, and CLS measures how much the page jumped while it loaded. TTFB carries no weight in the score; the other five do.

MetricPhoneDesktopWeight
FCPFirst Contentful Paintgreen up to 1.8 sred above 3 sgreen up to 934 msred above 1.6 s10%
LCPLargest Contentful Paintgreen up to 2.5 sred above 4 sgreen up to 1.2 sred above 2.4 s25%
TBTTotal Blocking Timegreen up to 200 msred above 600 msgreen up to 150 msred above 350 ms30%
CLSCumulative Layout Shiftgreen up to 0.10red above 0.25green up to 0.10red above 0.2525%
SISpeed Indexgreen up to 3.4 sred above 5.8 sgreen up to 1.3 sred above 2.3 s10%
TTFBTime to First Bytegreen up to 800 msred above 1.8 snot scored
Lighthouse 13's thresholds: a value at the green line scores 90 of 100, at the red line 50.

How is the grade calculated?

The grade is the median run's Lighthouse performance score: A from 90 (Google's green), B from 80, C from 65, D from 50 and F below 50. Each metric is scored on its own curve and weighted, and the result card shows the points each one cost.

We run three times because scores move from run to run, and show all three on a 0–100 track. One odd run never decides the letter. A page that never stops loading gets one or two runs instead, and its report says so.

Weights
TBT 30%LCP 25%CLS 25%FCP 10%SI 10%
Letters
F 0–49D 50–64C 65–79B 80–89A 90–100
How much each metric weighs in the score, and where each letter starts.

Why do PageSpeed Insights and GTmetrix show a different score?

Same Lighthouse, different servers, places and settings. With the Like Google PageSpeed setting we use the presets PageSpeed Insights runs (a simulated slow phone on 4G, or a desktop on cable) with our CPU slowdown calibrated so our servers score like Google's. Until a test server's slowdown is calibrated, its reports say it is provisional.

PageSpeed Insights runs once from a place you can't choose; we run three times from the place you pick. The Full speed setting loads the page at our server's own speed with no throttling, closer to GTmetrix's view, so its scores run higher.

The two settings on every report.Like Google PageSpeedFull speed
Network (phone)150 ms · 1.6 Mbps, simulatedthe test server's own
Network (desktop)40 ms · 10 Mbps, simulatedthe test server's own
Phone screenMoto G Power, 412×823Moto G Power, 412×823
CPUslowed, calibrated per servernot slowed
Runs3, median3, median
The two settings on every report.

Lab data and field data: which one counts?

The lab test is one fresh visit you can repeat after a fix: the same place, device and network each time, so a change shows the same day. Field data is 28 days of real Chrome users, the 75th percentile, from the Chrome UX Report.

Google's Core Web Vitals verdict comes from field data, and it can't move right after a fix: it averages four weeks. Use the lab to check a fix, and the field data to see it reach your visitors over the next month.

Lab (this test)

  • One visit, empty cache
  • A place, device and network you choose
  • Moves the moment you fix something
  • TBT stands in for interactivity

Field (real users)

  • Every Chrome visit that shares statistics
  • All their phones, networks and caches
  • The 75th percentile over 28 days
  • INP measures real interactions
The same page, two kinds of data.

How do I make my website faster?

Start with what costs the most points: the fixes are ordered by the saving the measurement proves. A slow server or a far one with no CDN delays everything after it; render-blocking files delay the first paint; heavy JavaScript blocks taps; a late lead image delays LCP.

Ads, analytics and other sites' scripts usually have to stay: delay them until the page is usable, or serve them yourself. Then test again: the report compares before and after and marks each earlier fix as fixed or still open.

FindingFixed in
The server is slow to answerslow-serverthe sitethe CDNthe server
No CDN: far places wait for one serverno-cdn-far-regionthe CDN
Render-blocking CSS and JavaScriptrender-blockingthe site
JavaScript keeps the main thread busymain-thread-busythe site
The largest paint comes latelcp-slowthe sitethe CDN
Images bigger than the screen needsimage-deliverythe sitethe CDN
Your CDN served an old copycdn-stale-copythe CDN
Common findings and where each is fixed.

Questions about the website speed test

Is the Website Speed Test free?

Yes. Every feature is free and needs no sign-up: tests from up to 3 places at once, phone and desktop, the filmstrip and the replay of the load, the waterfall, real users' data and the fixes. A visitor can start a few tests a minute, and the daily test time is shared fairly so the service stays free for everyone.

Why is my score different every time?

A page never loads exactly the same way twice: servers answer at different speeds, ads change, and the network varies. That's why we run Lighthouse three times per place and grade the median run, showing all three scores so you can see the spread (a page that never stops loading gets one or two runs instead, and its report says so). A difference of a few points between tests is normal.

Do you cache results?

Never for a new test. Every test starts a fresh browser with an empty profile, no cache and fresh DNS, so a fix you just made shows at once. If your own CDN serves a stored copy of the page, the report says so and how old it is, because real visitors get that copy too.

Which places can I test from?

Full browser tests run in 9 AWS regions: Virginia, Oregon, São Paulo, London, Frankfurt, Mumbai, Singapore, Tokyo and Sydney, up to 3 in one test. Each place keeps a test server running, so a test starts right away, or waits its turn behind another test there. Server timing always comes from our 11 places.

What do the Like Google PageSpeed and Full speed settings mean?

The Like Google PageSpeed setting uses the presets PageSpeed Insights runs: a simulated phone on slow 4G with its CPU slowed, or a desktop on a cable line, so scores compare with Google's. The Full speed setting loads the page at our test server's own speed with no throttling, which shows what a fast connection gets and scores higher.

Why did my site turn the test away?

Some sites use bot protection (DataDome, HUMAN, Cloudflare and others) that blocks automated browsers. Our test browser says what it is in its user agent, so it can be stopped. The report shows what the site sent back. If it's your site, allow our user agent or our test servers' fixed addresses, listed in the next answer.

Which IP addresses do your test servers use?

Every browser test comes from one of these fixed addresses, one for each of our 9 places, so you can allow them in a firewall or bot protection: 184.195.23.138 (Virginia), 54.201.75.231 (Oregon), 54.21.28.0 (São Paulo), 13.41.183.26 (London), 63.176.218.44 (Frankfurt), 15.206.239.6 (Mumbai), 13.229.17.166 (Singapore), 3.114.170.116 (Tokyo), 15.134.82.51 (Sydney). The server timing comes from Cloudflare's network instead, which has no fixed list. Our test browser also names itself in its user agent.

How long are reports kept?

Each report has its own link that you can share; it is kept with its screenshots for 30 days, then deleted. Report links are never listed or indexed, and we never publish the sites people test. The site you test sees our test server, never your address, and the AI that writes the fixes gets only the report.

Can I compare before and after a fix?

Yes. Press Test again on a report: the new test runs with the same page, places and settings, and the result shows every metric before and now. Each fix from the earlier test is marked fixed, improved or still open, judged from the new test's own data.

Missing something?Suggest a feature or a new tool

Related tools