Free VPN vs paid: same day, same line, seven tests
One free subscription and one paid subscription, benchmarked on the same machine over the same connection on the same day. Latency, bandwidth, packet loss, peak hours, streaming, AI tools, uptime. All numbers included.
This one is mostly numbers.
People ask how big the difference is, and "paid is better" is not an answer. So I ran two subscriptions on the same day, same machine, same broadband line, across seven measurements.
Conditions
- Date: 8 May 2026
- Line: 500 Mbps residential fibre
- Machine: Windows 11, Clash Verge Rev
- Free side: best-latency Japanese node from a public subscription pulled that morning
- Paid side: Luobo Cloud, also a Japanese route
- Method:
speedtest-cliagainst one Tokyo endpoint, three runs, median
Both sides pinned to a single node with auto-switching and load balancing disabled, so the only variable is the service.
Seven measurements
1. Latency
| Free | Paid | |
|---|---|---|
| Idle | 82 ms | 71 ms |
| Under load | 340 ms | 96 ms |
Idle they are close. Under load is where it separates — run a download and measure latency simultaneously, and the free node climbs to 340 ms while the paid one gains 25.
In practice that is the difference between browsing normally while something downloads, and watching a slideshow.
2. Bandwidth, 14:00
| Free | Paid | |
|---|---|---|
| Down | 62.4 Mbps | 218 Mbps |
| Up | 8.1 Mbps | 96 Mbps |
The free node is respectable at midday. 62 Mbps carries 4K.
Upload is the weak spot — that shows up when transferring files, on video calls, and uploading images to AI tools.
3. Bandwidth, 21:30
| Free | Paid | |
|---|---|---|
| Down | 1.2 Mbps | 214 Mbps |
| Up | 0.4 Mbps | 91 Mbps |
This is the single most important pair of numbers in the article.
The free node fell from 62.4 to 1.2 — a 98% collapse. The paid line barely moved.
For scale: 1.2 Mbps means images crawl in and 360p buffers.
4. Packet loss, 10-minute ping
| Free | Paid | |
|---|---|---|
| Daytime | 0.3% | 0% |
| Peak | 11.4% | 0.2% |
11% loss means pages need reloading, video calls fall apart, and streamed AI output breaks mid-sentence.
5. Streaming
| Free | Paid | |
|---|---|---|
| Netflix | Proxy error M7111-5059 | Plays, 4K |
| Disney+ | Would not play | Plays |
| YouTube 4K | Heavy stalling | Smooth |
I have been testing free nodes against streaming services for six months. Not once has one worked. Shared datacentre IPs are long since blacklisted.
6. AI tools
| Free | Paid | |
|---|---|---|
| Site loads | Yes | Yes |
| Login completes | On the fourth node | First attempt |
| 800-word response, unbroken | Failed 2 of 3 | 3 of 3 |
Expanded in Using ChatGPT over a free VPN.
7. Uptime
Left idle and connected:
| Free | Paid | |
|---|---|---|
| First disconnect | 1 h 47 m | None in 12 h |
| Disconnects in 6 h | 4 | 0 |
Three differences the numbers do not show
Maintenance
A free node dies and that is your problem. A paid node dies and the provider replaces it; you press refresh.
That is the actual purchase. Not bandwidth — accountability.
Time
I kept a log. Six months on free nodes averaged 30 minutes a week of hunting, testing and swapping. Two hours a month.
Four dollars against two hours. Work out your own hourly rate. Long version in Do free nodes actually save money?
Reliability at the moment it counts
Hardest to quantify, and the one that decided it for me.
In March I was sending a client an 800 MB file. The free node died at 80%. Re-sending took two hours; they waited overnight.
You never know when it will go, and it always goes at the worst possible time.
Do free nodes still have a role?
Yes, and I still use them — the role just changed:
- Free nodes: emergencies, quick lookups, testing, borrowed computers
- Paid line: anything that has to work
Both, in parallel, for about four dollars a month.
Price reference
The plan table for the service I used as the paid side, so you can run your own numbers:
| Plan | Monthly | Yearly | Data | Peak | Devices |
|---|---|---|---|---|---|
| 萝卜标准 300G 普通组 | ¥29 | ¥290 | 300 GB | 150Mbps | 4 |
| 萝卜畅用 600G 普通组 | ¥39 | ¥390 | 600 GB | 300Mbps | 6 |
| 萝卜高速 1000G 精品组 | ¥69 | ¥690 | 1000 GB | 600Mbps | 8 |
| 萝卜旗舰 1500G 精品组 | ¥99 | ¥990 | 1500 GB | 1000Mbps | 12 |
| 私人订制组 独享IP | ¥666 | ¥6666 | 2048 GB 起（月上限 6TB） | 按需 | 15 |
Annual billing works out to two months free. On the 300 GB entry tier I use roughly 180 GB a month with browsing, AI tools and occasional video — so it fits. If you watch a lot of 4K, start at 600 GB.
Watch the traffic multiplier: standard nodes count 1×, optimised nodes count 2× on the standard group and 1.5× on the premium group. Optimised routes are faster and consume the allowance quicker. Factor that in before choosing a tier.
Read next
- Free proxy nodes: where they come from
- Luobo Cloud after three months
- Do free nodes actually save money?
- How to actually test a free node
Questions people keep asking
Is the difference really that large?
Small during the day, enormous at peak. Free nodes carry no bandwidth guarantee and no quality-of-service management; a paid plan caps users per line and schedules around congestion. The widest gap I recorded was 1.2 Mbps against 214 Mbps at the same minute.
Could a free node ever beat a paid one?
Briefly, yes. A freshly published node that nobody has found yet can benchmark beautifully. The window is usually hours to a couple of days before it saturates. You cannot plan around a window you cannot predict.
Do paid services have bad days too?
Yes. Lines get blocked, datacentres have incidents, congestion happens. The difference is that someone is on the hook to fix it — dead nodes get replaced, blocked IPs rotated. When a free node dies, that is simply the end of it.
What is the cheapest paid option worth having?
Entry tiers on mainstream providers run around ¥29 a month, roughly four dollars. That buys 300 GB, four concurrent devices and a maintained subscription. Start monthly, never annual, until you have watched it through two weekends.