Test sequence

  1. Latency and request loss: the browser sends 24 uncached requests to the test route. The progress view updates as samples return.
  2. Download: after a small warm-up request, three parallel response streams transfer incompressible data for about ten seconds.
  3. Upload: after a warm-up, two parallel requests send generated binary payloads for about ten seconds.

How each metric is calculated

Ping

The median completed round-trip request time, in milliseconds. Median is used to reduce the influence of a single unusually slow sample.

Jitter

The mean absolute difference between consecutive latency samples. It indicates how consistently the connection responds.

Request loss

Timed-out or failed samples divided by the 24 attempts. Browsers cannot expose raw ICMP or UDP counters, so this is application-level request loss.

Throughput

Successfully transferred bits divided by elapsed measurement time, converted to megabits per second (Mbps).

Published quality thresholds

Internet connection quality thresholds by metric
MetricExcellentGoodFairPoor
Ping≤ 40 ms≤ 80 ms≤ 150 ms> 150 ms
Jitter≤ 10 ms≤ 20 ms≤ 35 ms> 35 ms
Request loss≤ 0.2%≤ 1%≤ 2.5%> 2.5%
Download≥ 100 Mbps≥ 25 Mbps≥ 8 Mbps< 8 Mbps
Upload≥ 25 Mbps≥ 10 Mbps≥ 3 Mbps< 3 Mbps

Why another test can show a different result

Two services may use different servers, routes, payload sizes, connection counts, warm-up periods and test durations. A nearby native app can also bypass browser overhead. For a useful comparison, test the same device, connection and time of day, close other traffic and run at least three tests.

Known limitations

  • Wi-Fi quality and device performance are part of the result.
  • A short test may not fully saturate very fast multi-gigabit lines.
  • VPNs, proxies, data savers and security software can alter traffic.
  • The nearest route is not necessarily the route used by every app.
  • Ratings are guidance, not a service-level certification.

Measurement references