Methodology
How speed.new measures
speed.new starts measuring the moment the page loads. Everything below runs in your browser against the nearest Cloudflare data centre — the same network that serves a large share of the web you use.
Latency & jitter
We warm up the connection, then send 12 tiny requests and use the browser's Resource Timing to measure the network round-trip only. Ping is the median; jitter is the mean change between consecutive pings.
Download & upload
We open 4 parallel streams (8–12 on fast links) of incompressible data. Each request doubles in size until it lasts about 1.5 seconds, so we measure your line rather than request overhead. The first 1.5 seconds are thrown away (TCP/QUIC ramp-up), throughput is sampled every 200 ms, and the headline is the 90th percentile of those samples — what your connection sustains at its best. The median is kept too. A phase ends when three one-second windows agree within ±5 %, or after 10 s (download) / 8 s (upload).
Latency under load & bufferbloat grade
While data is flowing we keep pinging every 250 ms. If latency balloons under load, video calls and games stutter even on "fast" connections. The grade uses the worst increase over idle latency:
| Grade | Increase | Meaning |
|---|---|---|
| A+ | < 5 ms | Perfect for gaming & calls |
| A | < 30 ms | Great for gaming & calls |
| B | < 60 ms | Mild bufferbloat |
| C | < 200 ms | Calls may stutter under load |
| D | < 400 ms | Heavy bufferbloat |
| F | ≥ 400 ms | Severe bufferbloat |
Data caps
A full test transfers at most 250 MB down and 100 MB up. On cellular, Save-Data or very slow links we run a light test (latency + up to 25 MB down) and offer the full test. Tests never start in a hidden or prerendered tab.
"Faster than X % of your neighbors"
Your download is compared with tests from the same ISP and city in the last 24 hours — only when at least 20 such tests exist. VPN, proxy, iCloud Private Relay and cloud-hosting networks are excluded from these comparisons.
Caveats
Wi-Fi, other devices on your network, background downloads, browser extensions and power-saving modes all affect results. Above ~1–2 Gbps some devices become the bottleneck — we'll tell you when that seems likely. For the truest number, test over Ethernet.