How lag.dev compares

Last updated 13 October 2026. Other tests' details are as they published them that day.

Every major speed test now measures some form of latency under load. They differ in where they measure, what they leave out and what they keep. Here is how lag.dev compares, including where the others do better.

At a glance

Five tests side by side, October 2026
Test	Measures	Where	Latency under load	Stability over time	Data per run	Privacy
fast.com (Netflix)	Download; upload and latency after "Show more info"	Netflix's own servers	Unloaded and loaded latency; loaded is one number, not split by direction	A warning when the network looks unstable	Not published; shows the data used, and lets you set duration and connections	We couldn't read a fast.com notice; it loads with JavaScript
Speedtest (Ookla)	Download, upload, idle and loaded latency, jitter; packet loss reportedly in its apps	Nearby servers from a network of over 15,000	Interquartile mean of pings during download and during upload	None	Users report 100–200 MB at about 100 Mbps	Its policy covers licensing results, precise location and behavioural advertising
Cloudflare	Download, upload, latency, jitter, loaded latency, packet loss; scores for streaming, gaming and calls	The nearest edge of its network in 330+ cities	From the download and upload phases	None	Grows with your speed; up to about 1.27 GB on multi-gigabit links, by our arithmetic on its open-source engine	Raw results downloadable; results published as open data through M-Lab
Waveform	Download, upload, idle and loaded latency, jitter; a letter grade	Cloudflare, Fly.io and Fastly servers	Mean increase under load, graded A+ (5 ms or less) to F (over 400 ms)	Disclaimed: it "measures latency under load, not reliability"	Not published	Not reviewed here
lag.dev	Latency to 8 servers at once, latency under load at each, download and upload per server	3 servers near you and 5 far, at the same time	p50 and p95 during each server's own load, and the increase over idle	Continuous pings to all 8 while the page is open; gaps and spikes over time	About 5 MB per Mbps of combined download and upload speed; capped at 2 GiB per server each way	Stores no IP address; keeps network and city-level location for 7 days; no analytics

Where the others do better

Speedtest has far more servers, so one is usually close to you, which makes it better at finding your line's maximum speed. It also has apps and a command-line tool.
Cloudflare measures packet loss, scores results for streaming, gaming and calls, lets you download the raw measurements, and publishes results as open data.
Waveform sums up bufferbloat in a single grade with published thresholds for gaming and calls.
fast.com is the simplest: one number from the servers Netflix streams from, with settings for a longer or heavier test.

Where lag.dev is different

It measures eight servers at once, three near and five far, so you see latency across regions, not just to the nearest server.
It keeps pinging, so it shows how steady latency stays over time, not a single burst.
Each load shows on every server's latency plot, so you see how your own connection's queue affects every path.
Its server stores no IP address.

It also has limits. It has no packet-loss test. Its speeds measure the path to each server, which is a URnetwork provider's machine, not your line's maximum. And it has far fewer servers than Speedtest. The methodology explains what each figure means.

Also worth knowing

Google's speed test, one click from a search for "internet speed test", is run by M-Lab. It measures download, upload and latency up to 700 Mbps. Google shares your IP address with M-Lab, which publishes results publicly.
Apple's networkQuality, built into macOS and iOS, reports responsiveness in round trips per minute (RPM), the measure the IETF's responsiveness draft defines.

Sources

Run the test