WireGuard, Hysteria2, VLESS or Trojan: Which Protocol Should You Use?
Your subscription offers five protocols and no guidance. Here is what each one is actually good at, why the fastest protocol on one network is the worst on another, and how to test instead of guess.
WireGuard, Hysteria2, VLESS or Trojan? 🔀
Open a modern proxy client and your subscription probably lists the same locations several times over, each with a different protocol tag. No explanation, no recommendation.
Here is the short answer up front: there is no universally best protocol. The one that wins on your home Wi-Fi can be the one that fails on hotel Wi-Fi, and the difference is usually the network between you and the server — not the protocol's theoretical quality.
What follows is what each one is genuinely good at, and how to find yours in ten minutes.
📦 The Field, Briefly
| Protocol | Transport | Where it shines | Where it struggles |
|---|---|---|---|
| WireGuard | UDP | Low overhead, fast handshakes, excellent on clean networks | Networks that throttle or block UDP |
| Hysteria2 | UDP (QUIC) | Lossy and long-distance links; aggressive congestion control | Same UDP problem; can be conspicuous |
| VLESS | TCP (often + TLS/Reality) | Blends into normal HTTPS traffic; survives restrictive networks | More per-connection overhead than UDP options |
| Trojan | TCP + TLS | Looks like ordinary HTTPS; simple and well-supported | Bounded by TCP behaviour on bad links |
| Shadowsocks | TCP/UDP | Lightweight, mature, widely supported | Older designs are easier to fingerprint |
Newer entries such as TUIC, ShadowTLS and AnyTLS show up in some subscriptions too, and the same logic applies: test them, do not assume.
🌐 The Real Variable Is UDP
Notice the pattern in that table. The single biggest factor is not the protocol's design — it is whether the network you are on treats UDP well.
WireGuard and Hysteria2 are UDP-based. On a clean connection, that is an advantage: less handshake overhead, faster recovery from packet loss, better behaviour on long-distance links. Hysteria2 in particular is built for links that lose packets.
But plenty of networks — corporate Wi-Fi, some hotels, some mobile carriers — throttle UDP, deprioritise it, or block everything except DNS. On those networks a UDP protocol can be dramatically slower than a TCP one, or fail outright.
TCP-based options (VLESS, Trojan) give that up in exchange for looking like ordinary HTTPS traffic, which is exactly what you want on a restrictive network.
This is why the answer changes when you change networks, and why a protocol that felt broken at the office can be the fastest one at home.
⚡ Speed Is Not a Property of the Protocol Alone
A protocol cannot make a congested server fast. In measurements we took across one subscription, throughput varied by more than sixty times between servers — and the variation did not track protocol type nearly as strongly as it tracked the individual server's condition.
So before concluding "Hysteria2 is faster than VLESS", make sure you are not actually comparing a healthy server to an overloaded one.
Two corollaries worth internalising:
- Latency does not predict throughput. The lowest-ms entry in a list can deliver the worst transfer speed. We measured a 488 ms server at 0.9 Mbps and a 587 ms server at 55.9 Mbps in the same sitting — the full numbers are here.
- A single test proves little. Server load changes by the hour.
🧪 How to Actually Find Yours
Ten minutes, done properly, beats any recommendation article — including this one.
- Pick one location and stay with it. Comparing a Japan VLESS entry against a Germany WireGuard entry tells you nothing about protocols.
- Test a real transfer, not a page load. Page loads are dominated by round trips and hide bandwidth differences. Download something substantial.
- Test each protocol at that location, back to back, within a few minutes.
- Repeat on your other networks. Home, office, mobile data. Write down which one wins where — this is the part people skip, and it is the part that matters.
- Re-check after a few weeks. Servers get reprovisioned; networks change policy.
You will typically end up with two answers: a preferred protocol for unrestricted networks, and a TCP fallback for the restrictive one you occasionally have to use.
🧰 Practical Rules of Thumb
- Home or mobile data, unrestricted: start with a UDP option (WireGuard, Hysteria2).
- Corporate, hotel, or campus Wi-Fi: start with a TCP+TLS option (VLESS, Trojan).
- Long-distance or lossy links: Hysteria2 is designed for exactly this.
- Nothing connects at all: the protocol is probably not the problem — check whether your traffic is even going through the tunnel.
🎛️ Let the Client Do Some of This
Manually testing every entry does not scale past a handful of servers. AI Booster measures per-node latency through its core and defaults to a selection strategy that ranks by health, so servers that cannot be reached from your current network are left out of the rotation instead of receiving a share of your traffic.
That handles "this server is dead right now" automatically. Choosing which protocol suits the networks you actually use is still worth ten minutes of your own testing.
Read more: node selection and what latency means · configuration reference
Which protocols appear depends on your subscription and client version — a provider decides what to offer, and the client decides what it can parse.