How to test a free node properly: four steps

Most people test a node exactly one way: press the latency button, pick the smallest number.

Roughly half of what that selects will not work.

Here is the full filter. Fifteen minutes to go from a hundred candidates to the two or three worth keeping.

Why latency alone is not enough

That number measures one thing: how long an HTTP request to a test endpoint took to come back.

It tells you the node is alive and the round trip is short.

It tells you nothing about:

• Throughput — whether it can fill your connection
• Stability — whether it drops three times a minute
• Packet loss — whether video stalls and streamed AI output breaks
• Peak-hour behaviour — the only period most people care about

I have used plenty of nodes that answered in 60 ms and delivered 300 KB/s. The route is open; there is simply no capacity on it.

Step 1: bulk latency, cut most of them (2 minutes)

Every Clash-family client tests a whole group at once.

Discard immediately:

A hundred nodes typically leaves twenty or thirty.

Step 2: measure real throughput (about a minute each)

The step that matters, and the one most people skip.

Switch to the node, then:

speedtest-cli --simple

Ping: 78.221 ms
Download: 62.40 Mbit/s
Upload: 8.13 Mbit/s

Or open fast.com in a browser.

My cutoffs:

Check upload too. Below 2 Mbps up, file transfers, video calls and image uploads to AI tools all become painful. Free nodes are consistently worse on upload than download.

Step 3: stability (10 minutes, run it in the background)

Throughput is meaningless if it cannot hold.

ping -t 1.1.1.1     # Windows
ping 1.1.1.1        # macOS / Linux

Ten minutes, then read the summary.

Five percent loss is invisible while browsing. It destroys video calls, and it will break a streamed AI response mid-sentence — which is exactly the complaint behind "the internet works but ChatGPT keeps cutting out".

Step 4: retest at peak (the decisive one)

Run steps 1–3 during the day, then retest the finalists between 8 and 10 pm.

Not optional. From my own log:

A 98% collapse at peak, on a node that measured beautifully six hours earlier.

You will use it in the evening. Judge it in the evening.

Use-specific tests

Passing the four steps means the node is basically sound. Whether it does your job is a separate question.

AI tools

Ask for 800 words and see whether it completes without breaking.

I have had nodes clear all four steps and fail this. Details: Using ChatGPT over a free VPN.

Streaming

Play actual 4K content. Ignore "unblock checker" pages — they test IP geolocation, not whether playback works.

Free nodes fail this essentially always; I have not had one succeed in six months.

Gaming

Jitter, not bandwidth. Latency bouncing between 80 and 90 ms is fine; 80 to 300 ms is unplayable.

Make it a routine

What I actually do:

1. New subscription → step 1, keep about twenty
2. Steps 2 and 3 on five of them → keep two or three
3. Evening retest → pick a primary and a backup
4. Repeat weekly (median free node lifespan: six days)

Fifteen minutes once you have done it a few times.

But notice the shape of that: it is a weekly chore. Fifty-two weeks at fifteen minutes is thirteen hours a year. That arithmetic is part of why I ended up paying — the full version is in Do free nodes actually save money?

Read next

• Free proxy nodes: where they come from
• Free VPN vs paid service: seven measurements
• Connected but nothing loads
• Is using free nodes safe?