It is slower than the speed I pay for.

Advertised figures are maximums, not delivered speeds. And over Wi-Fi you may be measuring the Wi-Fi rather than the line — try it wired.

The number changes every time.

Normal. Lines are shared, so time of day and congestion move it. Read the trend across several runs, not one figure.

A dedicated app reports higher.

Browser-based measurement has limits that native apps do not. Use this for comparing before and after under matched conditions.

What is it talking to?

Cloudflare’s measurement endpoint; not this site’s server. Each run transfers up to 21MB down and 6MB up — worth knowing on a metered connection.

Wi-Fi is slow. What do I try first?

Check which band you are on. 2.4 GHz reaches further but runs slower and collides with microwaves and the neighbours; 5 GHz is faster but stopped by walls. Re-run the test standing next to the router on 5 GHz — a large change means the bottleneck is the Wi-Fi rather than the line. After that, move metal and fish tanks out of the path.

It is only slow in the evening.

Shared capacity. Fibre in an apartment block, or a congested local exchange, sags at peak hours. Measure at the same time on several days: a consistent evening dip points at the provider’s equipment, while day-to-day randomness points back at your own Wi-Fi and hardware.

Can I do other things while it runs?

The result will come out lower. Streaming during a test takes bandwidth the test cannot use. For an accurate figure, stop other tabs and other devices on the same connection first. If what you want to know is how the line performs while in use, leave them running.

How fast you receive. It drives video and page loading. Even 4K streaming needs about 25Mbps, so past 100Mbps you rarely feel a difference.

How fast you send: video calls, posting photos and video, cloud sync. On some connections it is far narrower than the download figure.

Response time and jitter

Round-trip time and how steady it is. For calls and competitive games these matter more than raw speed — high jitter is what makes audio drop and inputs skip.

A factor of eight. Line speed is counted in bits per second, file size in bytes. A 100 Mbps connection tops out around 12.5 MB/s of actual download. "100 Mbps but only 12 MB/s" is the expected result, not a fault.

It depends on the job. Browsing and video calls want around 10 Mbps; 4K streaming around 25. Simultaneous users add up, so a household with four people streaming separately is comfortable at 100 Mbps. Raising the number only changes how things feel while you are below what you need.

Why does measuring cost data?

Speed can only be measured as "how long did a known amount of data take", so there is no way to do it without moving data — and more of it makes the result more accurate. This test uses up to 21 MB down and 6 MB up. On a metered connection that counts against your allowance.

Back to all tools／ Network

Internet speed test

This tool exchanges data directly with Cloudflare’s measurement endpoint (speed.cloudflare.com). A run transfers up to 21MB down and 6MB up, and the other end is not this site. What is sent is filler with no content — none of your files or input is involved.

Measures download and upload throughput, response time (ping) and jitter. The measurement runs against Cloudflare’s speed endpoints — it does not pass through this site’s server.

Measure

Press Measure for response time, then download, then upload. It takes ten to twenty seconds.

Output

Press Measure for response time, then download, then upload. It takes ten to twenty seconds.

Measurement runs against Cloudflare's test endpoints and never touches this site's own server. Jitter is reported alongside speed because calls and meetings depend on stability more than throughput.

All tools

Why this one

Connection setup is excluded: The first request carries connection and TLS setup. Dividing by total time makes fast lines look slow, so a throwaway request warms the connection before anything is timed.
Fast lines are not under-reported: One connection cannot saturate a fast link. Larger steps run several in parallel and are timed together, so hundreds of Mbps come out close to reality.
The transfer is capped: Testing “for five seconds” makes fast connections move enormous amounts of data. This is bounded by size instead: never more than 21MB down and 6MB up per run, and it stops early on slow lines.

How to use it

Press measure
Response time, then download, then upload. Ten to twenty seconds in total.
Keep conditions equal
Same place, same device, same time of day. Another device streaming video is enough to move the numbers.
Compare before and after
After changing a setting, measure again. The change reads better than any single figure.

Reading the numbers

Download: How fast you receive. It drives video and page loading. Even 4K streaming needs about 25Mbps, so past 100Mbps you rarely feel a difference.
Upload: How fast you send: video calls, posting photos and video, cloud sync. On some connections it is far narrower than the download figure.
Response time and jitter: Round-trip time and how steady it is. For calls and competitive games these matter more than raw speed — high jitter is what makes audio drop and inputs skip.
Mbps versus MB/s: A factor of eight. Line speed is counted in bits per second, file size in bytes. A 100 Mbps connection tops out around 12.5 MB/s of actual download. "100 Mbps but only 12 MB/s" is the expected result, not a fault.
How much is enough?: It depends on the job. Browsing and video calls want around 10 Mbps; 4K streaming around 25. Simultaneous users add up, so a household with four people streaming separately is comfortable at 100 Mbps. Raising the number only changes how things feel while you are below what you need.
Why does measuring cost data?: Speed can only be measured as "how long did a known amount of data take", so there is no way to do it without moving data — and more of it makes the result more accurate. This test uses up to 21 MB down and 6 MB up. On a metered connection that counts against your allowance.

Questions

It is slower than the speed I pay for.: Advertised figures are maximums, not delivered speeds. And over Wi-Fi you may be measuring the Wi-Fi rather than the line — try it wired.
The number changes every time.: Normal. Lines are shared, so time of day and congestion move it. Read the trend across several runs, not one figure.
A dedicated app reports higher.: Browser-based measurement has limits that native apps do not. Use this for comparing before and after under matched conditions.
What is it talking to?: Cloudflare’s measurement endpoint; not this site’s server. Each run transfers up to 21MB down and 6MB up — worth knowing on a metered connection.
Wi-Fi is slow. What do I try first?: Check which band you are on. 2.4 GHz reaches further but runs slower and collides with microwaves and the neighbours; 5 GHz is faster but stopped by walls. Re-run the test standing next to the router on 5 GHz — a large change means the bottleneck is the Wi-Fi rather than the line. After that, move metal and fish tanks out of the path.
It is only slow in the evening.: Shared capacity. Fibre in an apartment block, or a congested local exchange, sags at peak hours. Measure at the same time on several days: a consistent evening dip points at the provider’s equipment, while day-to-day randomness points back at your own Wi-Fi and hardware.
Can I do other things while it runs?: The result will come out lower. Streaming during a test takes bandwidth the test cannot use. For an accurate figure, stop other tabs and other devices on the same connection first. If what you want to know is how the line performs while in use, leave them running.

Measure

Output

Related tools

Why this one

How to use it

Press measure

Keep conditions equal

Compare before and after

Reading the numbers

Questions