What this speed test measures
Download and upload are measured by moving real bytes between your browser and Cloudflare's nearest edge, then dividing bits by seconds. The number you see is the 90th percentile of the individual round measurements, which is how Cloudflare's own tool reports it: a single slow chunk caused by a background app should not define your line.
Latency is the round trip to that same edge, reported as the median of 20 samples. Jitter is how much those samples differ from each other. Loaded latency is the same measurement taken while a transfer is running — the gap between the two is bufferbloat, and it is the number that explains why a video call falls apart the moment someone starts a download.
This is a measurement to the edge of a content network, not to a specific game server or website. It tells you what your access line can do; it cannot tell you that a particular server on the other side of the world is having a bad day.
What is a good internet speed in 2026?
The FCC raised the fixed broadband benchmark to 100 Mbps download and 20 Mbps upload in its 2024 Section 706 report (FCC 24-27, 14 March 2024) — four times the 25/3 Mbps standard that had stood since 2015. The same report sets a long-term goal of 1 Gbps down and 500 Mbps up, and a separate 35/3 Mbps benchmark for 5G mobile service.
In practice the honest answer depends on how many people share the line. One person streaming 4K needs 15 Mbps. A family of four with two 4K streams, a video call and a game console needs roughly 40–60 Mbps of headroom, not 1 Gbps. Upload matters more than most plans admit: it is what carries your camera in a call and your files to the cloud.
- Under 25 Mbps: fine for one person browsing and watching HD, tight for anything else.
- 100/20 Mbps: meets the current FCC broadband definition and covers a typical household.
- 500 Mbps and above: only pays off when several people transfer large files at the same time, or when your upload also scales.
How much speed do you actually need?
Netflix publishes its own numbers: 3 Mbps for HD 720p, 5 Mbps for Full HD 1080p and 15 Mbps for UHD 4K. Those are per stream, so two 4K televisions want 30 Mbps of download before anything else happens on the line.
Zoom asks for 1.2 Mbps for a one-to-one 720p call, 3.8 Mbps up and 3.0 Mbps down for 1080p, and 2.6 Mbps up for a 720p group call. Voice alone is only 60–80 kbps and screen sharing 50–150 kbps. Video calls are an upload problem far more often than a download problem.
Online games barely use bandwidth — a few hundred kilobits per second — but they are ruthless about latency and jitter. If your download is 500 Mbps and your game still rubber-bands, bandwidth was never the problem; check the ping and jitter numbers instead.
Why your result is lower than your plan
The most common cause is Wi-Fi, not the line. A 2.4 GHz channel shared with neighbours and microwaves rarely sustains more than 40–70 Mbps, and an older 802.11n device is capped around there no matter what you pay for. Move to 5 GHz or 6 GHz, or test once with a cable to separate the two.
- A 100BASE-TX port or an old cable caps you at about 94 Mbps of real throughput, which looks suspiciously like “my gigabit line gives 90 Mbps”.
- A VPN adds encryption overhead and a detour, typically costing 10–30 % and a lot of latency.
- Other devices matter: a cloud backup or a game update running in the background takes its share before this test starts.
- Evening congestion in your ISP’s local segment is real; the same line can differ by a factor of two between 3 a.m. and 9 p.m.
- A browser tab in the background is throttled by the browser itself, so keep this tab visible while the test runs.
Bufferbloat: why calls stutter on a fast line
Routers and modems keep large queues so that a burst of traffic is not dropped. When a download fills those queues, every other packet — your voice, your game input, a DNS lookup — has to wait behind megabytes of video. That waiting is bufferbloat, and it is why this page measures latency again while transferring.
An increase under 10 ms is nothing. 20–50 ms is where a call starts to sound clipped. Above 100 ms the line is unusable for real-time traffic while anyone is downloading, even at gigabit speeds.
The fix is on your router, not on your plan: enable SQM or fq_codel (OpenWrt, many ASUS and MikroTik models, Eero and Ubiquiti call it “smart queue”), and set the shaper to about 90 % of your measured upload and download. You lose a few percent of peak speed and get tens of milliseconds of latency back.
Wired vs Wi-Fi: how to test properly
Measure three times and the answer becomes obvious. First with an Ethernet cable straight into the router: that is what your line can do. Then on Wi-Fi standing next to the router: the difference is your adapter and the band. Then from where you actually sit: the difference from step two is walls and distance.
Close other tabs and pause cloud sync before each run, keep the tab in the foreground, and use the Full profile when you are comparing — the Quick profile is for a fast sanity check, not for a before-and-after experiment.
How to read these numbers
Speeds are in megabits per second (Mbps), file sizes in megabytes (MB). Divide by eight to convert: 100 Mbps is about 12.5 MB/s, so a 1 GB game update takes roughly 80 seconds at full speed. Storage and browsers talk in bytes; networks talk in bits, and that factor of eight explains most “my download is slower than my speed test” complaints.
Results move between runs and that is normal. Wireless conditions, other devices, the edge you land on and the time of day all shift the number by 10–20 %. Treat a single run as an estimate and a repeated pattern as evidence.
The very first chunk of a transfer is deliberately discarded. TCP starts slowly and TLS has to be negotiated; counting those milliseconds would understate every line, so the first 100 KB round only warms the connection up.