How does this browser speed test actually work?
We open four parallel HTTPS connections to Cloudflare's global speed endpoint and pull a large payload, measuring throughput once the TCP windows have ramped up. Upload uses a single fetch POST of a 25 MB random body so the server cannot compress it. Ping samples the same origin via HEAD ten times. We trim the first second and last 200 ms of each leg to avoid TCP slow-start and ramp-down skew, then average the steady-state portion. Everything runs inside your browser — there is no native client, no installed agent, and no third-party SDK calling home.
What is a good download and upload speed in 2026?
For a single user, 100 Mb/s down and 20 Mb/s up handles 4K streaming, video calls, and large game patches comfortably. For a household of four with simultaneous streams, aim for 300–500 Mb/s down and 50+ Mb/s up. Anything above 1 Gb/s is mostly headroom — most websites and CDNs cap individual flows long before that. Upload matters more than people think: video calls, cloud backups, and screen sharing all suffer when upload is below 10 Mb/s. If your plan promises 1 Gb/s but you measure 300 Mb/s on Wi-Fi, the bottleneck is the radio link, not the ISP.
Why is my measured speed lower than what my ISP advertises?
Wi-Fi is the usual culprit — 2.4 GHz tops out near 100 Mb/s in practice, and even Wi-Fi 6 on 5 GHz rarely sustains gigabit at range. Run the test from a wired Ethernet connection to isolate the link from the radio. Other suspects: an older router, a saturated CPU on a cheap ISP gateway, VPN overhead, browser throttling on battery, and shared neighborhood congestion in the evenings. If wired numbers match the plan but Wi-Fi does not, upgrade the AP or move closer.
How is this different from your latency and bufferbloat tests?
Speed test measures peak throughput on an otherwise idle link — the headline Mb/s number. Our /network/latency tool runs a continuous ping graph so you can spot transient lag spikes and Wi-Fi handoffs that a 30-second speed test never sees. /network/bufferbloat deliberately saturates the link while pinging, exposing the queue-management failure that turns video calls into slideshows during a download. Use all three together: speed for capacity, latency for stability, bufferbloat for behaviour under load.
Is any of my data sent anywhere?
The bytes themselves go to Cloudflare's public speed endpoint, the same one used by speed.cloudflare.com — that is unavoidable, because the test needs a remote peer to measure against. We do not log results on our server, do not require an account, and do not embed any analytics SDK inside the test page. Your IP is visible to Cloudflare for the duration of the requests, exactly as it would be when loading any website.
Why does ping fluctuate so much between runs?
Ping over the public internet is a stochastic process — every hop has a queue, and queues fill and drain unpredictably. A single sample is almost meaningless; we take ten and report jitter alongside the mean for that reason. If you want stability data rather than a snapshot, run /network/latency for a few minutes and watch the p99 column. Sustained jitter above 30 ms or any visible packet loss usually points to a Wi-Fi problem, not the ISP.