Cloudflare edge · anycast · nothing to install
Download, upload, ping, jitter and latency under load — measured in your browser in about 20 seconds, inside a hard 50 MB data ceiling.
Ready
0
Mbps
Press start · ~20 s · ≤50 MB
Idle. Nothing is measured until you press start.
—
PingMeter score · how it's calculated
Thresholds are per active stream. Run a test and each row grades itself against your real numbers.
Stored only in this browser (localStorage). Nothing is uploaded, and clearing your browser data deletes it permanently.
No tests yet. Your first result will appear here with a trend line, so you can prove to your ISP that evenings are slower than mornings.
Every number above comes from the process below — no estimates, no marketing math.
PingMeter sends empty requests (?bytes=0) to the nearest Cloudflare edge and times the round trip until response headers arrive. Two probes are discarded as warm-up so DNS, TCP and TLS handshakes never inflate the result. The reported ping is the median of the remaining probes; jitter is the mean absolute difference between consecutive probes.
A small probe estimates your line rate, then 1–8 parallel streams pull binary payloads sized to keep each request busy for roughly a second. Bytes are counted as they stream in, not when a request finishes, so the gauge is genuinely live. The first slice of the window (TCP ramp-up) is discarded and the sustained mean of the rest becomes your result.
Incompressible random payloads are POSTed to the same edge. Progress events drive the animation, but only fully completed requests count toward the final figure — browser upload progress reports bytes handed to the operating system, which over-states throughput on high-latency links.
While the pipe is busy, empty probes keep running. The gap between idle and loaded latency is bufferbloat: the reason a "fast" connection still turns video calls into slideshows. PingMeter grades the increase from A (under 5 ms) to F (200 ms or more).
| Phase | Allocation | Typical duration | Stops when |
|---|---|---|---|
| Latency probes | < 20 KB | 1–3 s | Probe count reached |
| Download | 33 MB (66%) | 8 s | Time is up or budget spent |
| Upload | 15 MB (30%) | 8 s | Time is up or budget spent |
| Edge metadata | < 1 KB | instant | Single request |
A 50 MB ceiling means a gigabit line is only saturated for a fraction of a second, so results above roughly 500 Mbps are labelled Medium or Low confidence rather than dressed up as precise. Wi-Fi, VPNs, browser extensions, other devices and CPU load all sit inside the measurement path. Test on Ethernet, close background downloads, and run three tests before arguing with your ISP.
Most "slow internet" complaints are measurement artefacts. Eliminate these first.
The five questions we get most. There are 24 in the full FAQ.
Advertised speeds are the maximum the access link can carry, usually measured on Ethernet to the ISP's own server. PingMeter measures the whole path from your browser, through your Wi-Fi, to Cloudflare's edge. Wi-Fi overhead, device limits, VPNs and evening congestion typically account for a 10–40% gap. A gap over 50% on a wired device is worth a support ticket.
At most the ceiling you select — 50 MB by default, split 33 MB download and 15 MB upload. On slow connections the time limit stops the test long before the ceiling, so a 3G test may use under 2 MB. The exact figure is shown as Data used in your report.
No. Results, settings and history live in your browser's localStorage and never reach a PingMeter server — there is no PingMeter server. Cloudflare necessarily sees your IP address to serve the measurement requests, exactly as it would for any website. See the privacy policy.
Under 40 ms is comfortable, under 20 ms is excellent, and jitter under 10 ms matters more than the raw number. Loaded latency matters too: if ping triples while someone else streams, your router needs smart queue management. Details in the speed guide.
Because the data ceiling ran out before a long measurement window was possible — typical on gigabit links. Choose the Deep profile, or accept the figure as a lower bound. PingMeter would rather flag uncertainty than invent precision.
Compare it against real requirements for streaming, gaming, video calls and remote work.