ProtonVPN vs Mullvad: What We Measured

In short

Measured from one rented server in a datacentre (Kamatera, Toronto), through ProtonVPN in NL and Mullvad in NL, minutes apart, every six hours; the last cycle was aggregated 4 hours ago.

• Download speed: being measured again - the note below says why.
• Latency: about the same (91.6 ms against 90.4 ms), inside what the lab can tell apart.
• Time to connect: Mullvad ahead, 7.5 s against 13.0 s (42% less).

Speed through the tunnel is being measured again from 5 October 2026, 22:08 UTC. Before that, a TCP setting on our own machine capped every speed through the tunnel at about 275 Mbps, and both providers’ best runs stopped there: a limit of our stand, not of either provider. The figures come back once the new series has 10 cycles, 5 to go. Latency, time to connect, exits and DNS do not depend on that setting and are shown in full.

Side by side

Cycle by cycle

Both providers run in the same cycle, minutes apart on the same line, so each cycle is one like-for-like pairing. A median says where the middle is; this counts how often each came out ahead, and by how much in the middle cycle.

• Latency: Mullvad lower in 85 of 114 cycles, ProtonVPN in 29; the median gap per cycle was 1.0 ms, smaller than the lab treats as a difference.
• Time to connect: Mullvad quicker in 84 of 114 cycles, ProtonVPN in 19, level in 11; the median gap per cycle was 5.0 s.

Speed is counted only over cycles in one speed series. Time to connect is read in whole seconds every few seconds, so “level” means within one reading.

Whose network your traffic leaves from

Every run records the network the exit address is registered to. That is all an address says about itself: whether a provider owns that network or rents space in it is not something the stand can see, so this page does not claim either.

ProtonVPN

• Netid (AS49981) in 69 of 134 runs
• NForce Entertainment B.V (AS43350) in 65 of 134 runs

Mullvad

• 31173 Services Netherlands (AS39351) in 67 of 118 runs
• DataCamp Limited (AS212238) in 43 of 118 runs
• Mullvad VPN AB (AS43357) in 8 of 118 runs

DNS followed the exit: the resolvers that answered our lookups sat in the same network as the exit address in 99% of ProtonVPN's measured runs and 100% of Mullvad's. That means the lookups were answered inside the network the traffic left from, not by an outside resolver - measured against our own nameserver. Whether the provider or the network’s owner runs that resolver is not something the stand can see.

What this comparison cannot tell you

• Your speed. A datacentre line is faster than a home one; compare the two columns with each other, not with your connection.
• Either provider’s top speed. Each transfer is one connection over one route from our server, so the speed rows rank the two against each other. For the stand’s first month our own TCP setting capped that connection at about 275 Mbps, and both providers’ best runs stopped there.
• Other countries. Each side is measured through servers in NL; another location can rank differently.
• Streaming. The stand does not test whether either unblocks a service; the proxy flag above is the closest it gets.
• Privacy policy, logging and audits. A measurement shows how a connection behaves, not what a company does with it.
• IPv6 leaks. The stand's server has no IPv6 of its own, so it cannot test them yet.

Check the one you use

These figures describe the providers on our machine. What matters for you is your own connection: the leak test checks IPv4, IPv6, DNS and WebRTC with your VPN on.

🔍 ProtonVPN leak test 🔍 Mullvad leak test 🧪 How the lab measures