Getting Past Browser Speed Tests Without the Drama
I spent years watching engineers run browser-based speed tests and then complain when the numbers didn't match what their equipment was actually delivering. The gap between what a web test shows and what's happening on your network is usually massive, and it's almost never a software bug. It's just the inherent limitations of running a throughput benchmark inside a sandboxed JavaScript environment that shares CPU cycles with half a dozen other tabs. The command-line speedtest from Ookla fixes most of that. It runs as a native binary, talks directly to the nearest test server, and gives you measurements that are closer to what your actual infrastructure sees. You stop getting results that are 30 percent lower than your contract speed for no clear reason.
Unlock Your Internet Speed Potential With Speedtest Cli Test Internet Speed From Command Line
Download the binary from the Ookla website. The official source is speedtest.net/cli. They have packages for Linux, Windows, and macOS. On Linux, the deb and rpm files work cleanly. I usually grab the .deb version and run a basic install. Windows users get an MSI installer. macOS needs the dmg. Pick whichever matches your system. No one builds these things around package managers, so you're managing the install yourself. Once installed, you run it from the terminal. The default command is just speedtest. It will automatically select the closest server and run the full test sequence. That means latency, download, and upload. The output gives you a server name, ping, download speed, and upload speed. Pretty straightforward. Where it gets useful is when you start adding flags. The --server-id flag lets you pin a specific test server instead of relying on Ookla's automatic selection. Automatic selection is fine for general checks, but if you're troubleshooting a route to a particular PoP or CDN edge, pinning a server gives you repeatable results. You can list available servers with --list and filter by city or provider. The server database is large enough that you'll usually find something close to where your traffic actually goes.
I hit a specific issue once where the CLI kept resolving to a test server three hops away from my actual upstream provider's network. The auto-selected server was sitting on a completely different peering partner, and the latency numbers looked terrible while my actual user experience was fine. I found the right server by running --list with a city filter, then manually picking the one that shared the same ASN as my ISP. After that, my test results matched what the network team was seeing on the routers. Took about ten minutes to sort out instead of two days of back-and-forth with a vendor. The --format flag changes how output looks. json, csv, and plain text are the main options. JSON is the one I use most because it plays nicely with scripting. If you're running automated checks every hour and need to log results somewhere, parsing the plain text output is possible but unnecessarily painful. JSON gives you structured data without the regex wrestling. One thing people miss about the CLI is that it respects your system's DNS and routing. That means if your ISP has a misconfigured resolver or a weird default route, the CLI will surface it faster than a browser test ever would. Browser tests often go through their own DNS or get routed differently because they're hitting a web page first. The CLI skips that middle layer. It connects straight to the test server's IP using whatever routing decisions your OS makes.
Get the Full Details
There's also the --upload and --download flags if you only care about one direction. Useful when you're isolating whether a problem is inbound or outbound. You can also use --timeout to adjust how long the test waits before bailing. The default is usually fine, but on high-latency links like satellite connections, raising the timeout prevents premature failures. I should mention the limitations because people tend to treat this like it's the final word on bandwidth. It's not. The CLI measures throughput between your device and Ookla's test server at a single point in time. It doesn't tell you about congestion during peak hours, it doesn't measure packet loss across multiple paths, and it won't catch intermittent issues that only happen when your actual workload is running. If you need comprehensive monitoring, you'd want to pair this with something that does sustained testing over longer windows, maybe with multiple concurrent streams. The CLI is a snapshot tool, not a continuous monitor. Another thing to watch out for is that the test server selection still depends on Ookla's infrastructure, not yours. If you're behind a corporate firewall or a complex NAT setup, some test servers might not be reachable at all. I've seen this on networks with strict egress filtering where half the server list comes back as unreachable. In those cases, you need to work with the network team to whitelist the test servers or find a server that's accessible through your permitted routes.
Running the test from behind a VPN will also skew your results, obviously. The VPN adds its own overhead and may route you through a different exit node than your real traffic uses. If you're testing for troubleshooting purposes, run it both with and without the VPN to establish a baseline. The difference between the two tells you how much the VPN is costing you in performance. For most people, the default run is enough. Install it, type speedtest, and you'll get numbers that are more accurate than anything a browser can give you. When you need repeatability, automation, or direction-specific data, the flags open up without making the tool complicated. It's not going to replace a proper network monitoring platform, but for checking whether your connection is actually performing at contract speed, it does the job without the noise.
