Free proxy nodes: where to get them and how to test them

This is one of the two spines of this site. It runs long, because handing you a link is useless — the link dies on Thursday. What you need is a method for finding, testing and judging them yourself.

What a "free node" actually is

A node is one line of connection information you paste into a proxy client:

vmess://eyJhZGQiOiJleGFtcGxlLmNvbSIsInBvcnQiOiI0NDMi...
ss://[email protected]:8388#node-name
trojan://[email protected]:443?sni=example.com#node-name

Server address, port, cipher, password — encoded. Feed it to Clash, sing-box or Shadowrocket and the client knows what to do.

A subscription is a URL holding a batch of them, which the client re-fetches on a schedule:

https://example.com/api/v1/client/subscribe?token=xxxxxxxx

Subscriptions beat individual nodes, because when a node dies the provider replaces it and you just hit refresh. Look for subscriptions, not single links.

Four sources, ranked by how much they can be trusted

1. Trial allowances from paid providers — the best "free" there is

Plenty of paid services hand out a free tier or a trial, usually a few gigabytes. Try this category first, because:

• The nodes are the same ones paying customers get (sometimes rate-limited)
• You get a proper subscription URL that every client understands
• It will not vanish overnight; it runs until the data does

The catch is the size — typically 1 to 10 GB, then you either pay or move on.

2. Aggregator repositories on GitHub

A whole genre of repository does the same job: scrape publicly shared nodes, deduplicate, health-check, republish as a subscription, several times a day.

Search terms that find them: free node, v2ray subscribe, clash subscription, plus date-stamped repository names.

Quality: poor. Much of it is leaked subscription content from paid providers, which gets rotated out fast. Getting three to five usable nodes out of a hundred is a normal result.

3. Telegram channels

Many channels post nodes daily. Fastest updates, most variable quality.

4. Build your own

Not free exactly, but it can be very cheap — free-tier cloud instances, student credits, promotional machines.

You get a server nobody else is sharing and nobody else is watching. You pay in setup effort, and in having to move when the IP gets blocked. Pick a protocol first: SS vs VMess vs Trojan vs Hysteria2.

The fifteen-minute filter

Do not start using a batch of nodes. Run them through this first.

Step 1: bulk latency test

Every Clash-family client has a test-all button for a group.

Drop immediately:

• Timeouts
• Anything over 800 ms
• Anything showing 0 ms — that number is fake, the test endpoint got intercepted locally

A hundred nodes usually leaves twenty or thirty.

Step 2: measure real bandwidth, not latency

Low latency does not mean fast. This is the mistake everyone makes first. Plenty of nodes answer in 80 ms and then deliver 500 Kbps, because four hundred people are on them.

Switch to the node, then:

speedtest-cli --simple

Or open fast.com. Confirm the traffic is actually going through the proxy first — check an IP lookup page and make sure it shows the node's region. A surprising number of people benchmark their own home connection by accident.

My cutoff: under 5 Mbps down, discard. Ordinary browsing stutters below that.

Step 3: stability, the step almost everyone skips

Bandwidth means nothing if it cannot hold. Leave a ping running for ten minutes:

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

Two things matter: packet loss and jitter. Latency bouncing between 80 and 90 ms is fine. Between 80 and 600 ms means video will stall and streamed AI output will break mid-sentence.

Step 4: re-test in the evening

This one came from experience. A node that measured beautifully at 2 pm can be at a tenth of that by 9 pm. You will be using it in the evening, so judge it in the evening.

Full method: How to actually test a free node.

Lifespan data

47 nodes tracked, February to August:

Overall median: six days. Which means, on average, one maintenance session a week.

Three honest things about safety

1. The operator sees your metadata. HTTPS covers the contents. It does not cover which domain, at what time, for how long, and how much data.
2. Keep important accounts off free nodes. Banking, work mail, payments, government services. This is basic risk management, not scaremongering.
3. Configure routing rules. Domestic traffic direct, foreign traffic proxied. Faster, cheaper, and it stops you exposing things that did not need exposing.

Longer version: Is using free nodes safe?

When to stop optimising

Two or more of these and free is costing you more than it saves:

• More than 30 minutes a week spent swapping nodes
• You have already missed something because of a dropped connection
• You need AI tools that stream long responses without breaking
• You want 1080p or better
• Several devices at once
• A stable IP, or a specific exit region

I gave in on the second one. The reasoning is in Luobo Cloud after three months.

Read next

• Public subscription links that still update
• Free Clash nodes: get them and import them
• How to actually test a free node
• No nodes after importing a subscription