SwiftCheck

The problem

Speed tests report megabits per second. Operating systems, browsers and torrent clients report megabytes per second. The two differ by a factor of eight, so a connection measured at 10 Mbps downloads at 1.25 MB/s — and people reasonably conclude something is broken. Running the test also usually means visiting an ad-heavy site.

The approach

SwiftCheck reports every figure in MB/s, the same unit the file manager uses, and runs from the browser toolbar instead of a website. Throughput is measured with PerformanceResourceTiming — the browser's own network-stack timing — rather than a JavaScript wall clock, so the reading reflects bytes on the wire and is not skewed by script execution.

What it does

MB/s throughout

Real megabytes, matching what the browser's download manager reports. 100 Mbps displays as 12.5.

Adaptive test size

A 256 KB probe estimates the connection first, then the main transfer is sized to roughly three seconds at that speed — accurate on slow links without wasting data on fast ones.

Dynamic dial scale

The speedometer ceiling is set from the probe result, so a 1 MB/s connection gets a 0–3 scale with full needle travel instead of a needle pinned near zero.

Hardware-level timing

Uses responseStart to responseEnd from the browser network stack, measured at the OS level, rather than timing the fetch in JavaScript.

No tracking

No analytics, no data collection, no accounts. One outbound host: speed.cloudflare.com.

Pricing

Free and open source under the MIT licence.