Method comparison
There is no evidence-backed “most accurate” speed test for every purpose
Different tests can all be working correctly while producing different numbers. The useful question is which endpoint, traffic pattern, metric, and data policy match the question you are trying to answer.
Provider documentation reviewed July 10, 2026.
Documented differences
| Test | Published method or scope | Useful when | Data-policy note |
|---|---|---|---|
| WiFiSpeedTest | Eight parallel download workers, six upload streams, fixed sampling windows, trimmed 500 ms slices, and HTTP RTT to this site's Cloudflare edge. | You want this site's exact, published browser-to-edge method and browser-local history. | See the site's current data-flow disclosure. |
| FAST.com | FAST.com says it downloads from and uploads to Netflix servers; “Show more info” exposes upload plus loaded and unloaded latency. | You want a measurement using Netflix-operated test infrastructure. | FAST.com links to Netflix's privacy statement; review it before testing. |
| Cloudflare Speed Test | Cloudflare's published design article describes progressive payload sizes, repeated download/upload samples, latency and jitter, detailed statistics, and downloadable results. | You want Cloudflare's own test UI and its per-measurement detail. | Cloudflare's article says performance data is collected and anonymized under its privacy policy. |
| M-Lab NDT | M-Lab documents NDT as a single-stream bulk-transport capacity test reporting download, upload, and latency. | You need a documented research-oriented measurement whose published dataset can be queried. | M-Lab says the public IP and results are collected and test results are published. |
Choose based on the question
- “What did this browser achieve to this site?” Use WiFiSpeedTest and retain its method and conditions with the result.
- “What does Netflix's test infrastructure observe?” Use FAST.com.
- “Can I inspect multiple payload sizes and Cloudflare's detailed output?” Use Cloudflare's test.
- “Do I need a public research dataset?” Read M-Lab's collection policy first, then use NDT if publication is acceptable.
A lower result from one service does not, by itself, prove ISP throttling. Endpoints, routes, concurrency, protocols, payload sizes, server load, device state, and the aggregation formula must be controlled before attributing a cause.
Run a reproducible comparison
- Use the same device, browser, connection type, and physical location.
- Pause avoidable background traffic.
- Run each tool several times in alternating order.
- Record the time, endpoint, and tool version or method page.
- Compare medians and the spread, not only the highest number.