Commands tested on macOS 27.0.1 on 2026-10-20. The flags below come from man networkQuality on that release. Older guides show output that no longer matches, so trust the man page on your own Mac.
Run it: networkQuality in Terminal
networkQuality needs macOS 12 Monterey or later, as Apple’s support page says. It is already installed at /usr/bin/networkQuality.
- Open Terminal (Applications › Utilities).
- Type
networkQualityand press Return. - Wait for the summary. By default it tests upload and download at the same time. Our run took 45 seconds.
The man page warns that the test “will use data on your Internet service plan”. Our run moved 97.5 MB up and 45.9 MB down, so think twice on a phone hotspot.
The name has a capital Q, but on the standard case-insensitive APFS startup disk networkquality runs the same program.
Read the output: capacity, responsiveness (RPM) and idle latency
This is the summary from networkQuality -v on macOS 27.0.1, on a slow Wi-Fi connection:
==== SUMMARY ====
Uplink capacity: 18.993 Mbps
Downlink capacity: 8.787 Mbps
Responsiveness: Low (1.284 seconds | 46 RPM)
Idle Latency: 33.998 milliseconds | 1764 RPM
| Line | What it means | Unit |
|---|---|---|
| Uplink capacity | The most the connection carried from your Mac during the test (upload speed) | Mbps |
| Downlink capacity | The same toward your Mac (download speed) | Mbps |
| Responsiveness | How many round trips the connection completes per minute while it is busy, with a Low, Medium or High label | RPM, plus the delay in seconds or ms |
| Idle Latency | The round-trip time with nothing else on the line | ms, plus the same in RPM |
RPM is a delay turned upside down. The IETF draft that describes the test defines it as 60,000 divided by the round-trip time in milliseconds, so higher is better. Check it against the run above: 60,000 / 1,284 ms = 46.7, shown as 46 RPM; 60,000 / 33.998 ms = 1,764.8, shown as 1764 RPM.
The useful reading is the gap between the two latency lines: 34 ms idle against 1.3 seconds busy, about 38 times longer. A speed figure never shows that, and it is what makes a call freeze while a backup runs.
Apple’s page describes the labels in words, not numbers. Low: a movie download or an iCloud photo backup on the same network can make FaceTime calls or games unreliable. Medium: brief pauses or freezes when several devices share the network. High: apps stay well connected however many devices share it.
Two things to know when you compare with other guides:
- The example output in the server project the man page points to shows
Upload capacity,Download capacityandResponsiveness: High (1825 RPM). On macOS 27.0.1 the lines readUplinkandDownlink, responsiveness shows the delay next to the RPM, and idle latency has its own line. - One run is a sample. Here upload beat download, which is unusual for a home line. The
-voutput gives each figure anAccuracyofHigh,MediumorLow; this responsiveness figure hadAccuracy: Low, so run the test again, with other downloads stopped, before you act on it.
Flags that matter on macOS 27
networkQuality -h lists every option. These are the ones worth knowing, as man networkQuality describes them on macOS 27.0.1:
| Flag | What it does |
|---|---|
-v |
Verbose output: per-direction detail, accuracy, protocols, test server, interface and duration |
-s |
Tests download and upload one after the other instead of together |
-d / -u |
Skips the download or the upload test (both imply -s) |
-I en0 |
Binds the test to one network interface |
-c |
Prints machine-readable JSON, to the screen or to a file name you give |
-M 20 |
Sets the maximum run time, in seconds |
-p |
Runs the test through iCloud Private Relay |
-f h1, h2, h3, L4S, noL4S |
Forces an HTTP version, or turns L4S on or off |
-C url, -k |
Tests against your own server; -k skips certificate checks |
-S port |
Runs a networkQuality server on this Mac |
See which direction slows down: -s
The default run loads both directions at once and reports one responsiveness figure. With -s, the JSON output (-c) gains separate dl_responsiveness and ul_responsiveness values, according to the man page. Use it when calls break up and you want to know whether uploads or downloads cause it:
networkQuality -s -c
Test Wi-Fi or Ethernet only: -I
Find the interface your Mac uses now:
route -n get default | grep interface
interface: en0
networksetup -listallhardwareports maps names to ports. On this Mac, en0 is Wi-Fi:
Hardware Port: Wi-Fi
Device: en0
Ethernet Address: aa:bb:cc:11:22:33
Then run networkQuality -I en0. Test over a cable and over Wi-Fi in turn to see whether Wi-Fi is the limit.
Save results for a script: -c
networkQuality -c prints JSON instead of text. The man page lists the fields; the main ones are dl_throughput and ul_throughput in bits per second (not Mbps), base_rtt (idle latency in ms) and responsiveness in RPM. If a run cannot finish, the JSON includes error_domain and error_code.
Speed between two Macs: -S
networkQuality -S 8080 starts a test server on this Mac and prints the URL of its config file. On the other Mac, pass that URL with -C and add -k, because the certificate is self-signed. We did not run it for this page, since it opens a listening port. For speed inside your network, see testing speed between two computers and iperf3.
“command not found” and other errors
| What you see | Likely cause | What to do |
|---|---|---|
zsh: command not found: networkQualty |
A typo | Retype it: networkQuality |
command not found with the right spelling |
macOS older than Monterey | Check with sw_vers -productVersion; the tool needs macOS 12 or later |
command not found on Monterey or later |
Your PATH no longer includes /usr/bin |
which networkQuality should print /usr/bin/networkQuality; if it prints nothing, run that full path |
networkQuality vs speedtest-cli vs a browser test
“speedtest cli” can mean two different programs. Facts checked on 2026-10-20:
| networkQuality | Ookla Speedtest CLI (speedtest) |
speedtest-cli (Python) | speed.cloudflare.com | |
|---|---|---|---|---|
| Install | Built into macOS 12 and later | Homebrew: brew tap teamookla/speedtest, then brew install speedtest; version 1.2.0 for macOS |
pip install speedtest-cli or Homebrew |
None, runs in the browser |
| Servers | Apple’s (our run hit an aaplimg.com host, a domain registered to Apple) |
Ookla’s Speedtest server network | speedtest.net servers | Cloudflare |
| Latency under load | Yes, as responsiveness in RPM | Not stated; Ookla lists latency and packet loss | No; its README says latency only picks the nearest server | Yes, loaded latency |
| Output for scripts | Text or JSON | CSV, JSONL or JSON | CSV or JSON | A web page |
| State | Ships with macOS | Maintained by Ookla | Repository archived on 2026-04-30; Homebrew marks the formula deprecated as unmaintained | Live |
Where each one is the better choice:
- Ookla’s CLI when you want results from the same server network as the Speedtest website, or when you need packet loss.
- A browser test when you cannot use Terminal. Cloudflare says its test “doesn’t try to saturate your connection”, so it does not measure maximum throughput.
- networkQuality when you want a result with nothing installed, and when the question is “why does it lag when it is busy” rather than “how many megabits”.
- Avoid the Python speedtest-cli for new scripts. Its README warns it can report results “inconsistent with Speedtest.net”, and Homebrew will disable its formula on 2027-01-18.
Each uses its own servers, so the numbers will not match. Compare runs of one tool, not across tools.
Responsiveness and bufferbloat
A big gap between idle latency and loaded responsiveness has a name: bufferbloat. A router or modem keeps a long queue of packets while the line is full, and everything else, your call included, waits behind it. Our run, 34 ms idle and 1.3 seconds loaded, looks like a bad case. It ran over Wi-Fi, though, which adds delay of its own, so test over a cable before you blame the router.
Three things to know before you fix it:
- A faster plan rarely helps. The fix Apple suggests is Smart Queue Management (SQM) on the router, often listed as fq_codel or cake.
- Upload is the usual culprit, because it is the smaller direction and fills first. Run
networkQuality -s -cto see the two directions apart. - Run the test again after any router change. Expect responsiveness to improve, even if the speeds drop slightly.
For what counts as a good delay in milliseconds, see latency.
The same test with history, in an app
Speed Test in Netutilus does the same job in a window. Open Speed Test under Diagnostics in the sidebar and click Start. It shows download, upload, latency and jitter, a Responsiveness figure in RPM, and a bufferbloat grade from A+ to F set by how much the delay rose while the line was loaded. A grade of C or worse comes with the router fix spelled out. Speed Test is free.
History is part of Netutilus Pro. It runs the test on a schedule (Run automatically), can skip metered connections, and exports a CSV or an ISP report (PNG) to attach to a support ticket. networkQuality keeps no such record.
It is not the same measurement, so do not expect the same numbers:
- Netutilus tests against Cloudflare, with a second server only if the first refuses. networkQuality tests against Apple’s servers.
- Netutilus works out RPM from the median TCP connection time while the line is loaded, and shows the worse of the two directions. networkQuality measures transport, security (TLS) and HTTP round trips, which
-vlists one by one. - A Netutilus test uses up to about 1 GB, usually much less, because each half stops once the reading settles.
networkQuality stays the better tool for JSON in a script, a chosen interface or a server on a second Mac.