The numbers · October 2026
All the numbers
Pick a dataset and read one table at a time. Tap a column heading to sort it.
App benchmark
Open an app's name for all its numbers on one page.
Mac lab: 24 apps, 2 rounds
Each app tunnels traffic to a test server on the same Mac, so no network is in the way and the numbers show the app's own cost. Ranked by speed (30%) and energy efficiency (70%).
Scroll the table sideways for every measure.
| Speed, Gbit/s | At 200 Mbit/s | Build | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| App | Scoreof 100 | Average | Upload1 connection | Download1 connection | Upload4 connections | Download4 connections | EnergyJ per GB | CPUs per GB | MemoryMB | Core | MTU | Version |
| Throne | 93.9 | 16.9 | 13.5 | 18.9 | 17.1 | 18.7 | 0.288 | 5.11 | 30.5 | sing-box and Xray | 4000 | 1.4.0-beta.3 |
| sing-box CLI | 82.4 | 16.3 | 13.6 | 18.9 | 14.8 | 18.7 | 0.342 | 4.82 | 29.4 | sing-box | 4000 | 1.15.0-alpha.12 |
| SFM | 81.3 | 9.58 | 14.3 | 6.44 | 12.7 | 7.25 | 0.277 | 4.95 | 36.7 | sing-box | 4064 | 1.15.0-alpha.12 |
| Egern | 64.2 | 3.87 | 3.4 | 4.42 | 3.38 | 4.4 | 0.263 | 4.25 | 6.67 | Own core | 4000 | 2.20.0 |
| Surge | 57.9 | 10.2 | 11.8 | 9.28 | 12.5 | 7.96 | 0.463 | 8.66 | 201 | Own core | 4000 | 6.10.0 |
| Hiddify | 49.1 | 9.64 | 9.94 | 8.97 | 8.24 | 11.8 | 0.571 | 8.59 | 191 | sing-box | 9000 | 5.0.0 |
| FlClash | 48.4 | 10.5 | 11.4 | 9.44 | 11.1 | 10.2 | 0.605 | 11 | 25.6 | mihomo | 4000 | 0.9.1 |
| v2rayN | 46.2 | 8.69 | 7.38 | 11.1 | 7.05 | 10.1 | 0.597 | 10.9 | 58.9 | sing-box and Xray | 4000 | 7.25.6 |
| Clash Verge Rev | 45.7 | 7.72 | 8.23 | 8.6 | 6.54 | 7.68 | 0.576 | 10.6 | 27.3 | mihomo | 4000 | 2.5.8 |
| Koala Clash | 45.6 | 10.6 | 11.5 | 9.6 | 11.2 | 10.1 | 0.662 | 10.4 | 25 | mihomo | 4000 | 1.5.0 |
| Anywhere | 36.1 | 6.5 | 5.95 | 4.4 | 7.17 | 9.51 | 0.749 | 13.1 | 8.09 | Own core | 4064 | 2.4 |
| Shadowrocket | 31.8 | 1.96 | 1.47 | 2.37 | 1.54 | 2.76 | 0.537 | 9.61 | 7.19 | Own core | 4064 | 2.2.92 |
| Karing | 22.3 | 2.44 | 2.51 | 2.42 | 2.57 | 2.28 | 0.98 | 16.8 | 16.2 | sing-box | 4000 | 1.2.1 |
| HyperOrbit | 19.8 | 0.712 | 0.507 | 0.983 | 0.544 | 0.949 | 0.686 | 12 | 7.39 | Xray | 4000 | 1.5.4 |
| Rabbit Hole | 18.9 | 2.83 | 4.21 | 1.09 | 4.13 | 3.39 | 1.32 | 23.7 | 21.3 | mihomo | 4000 | 1.6.0 |
| Clash (Hako) | 16.2 | 2.33 | 3.16 | 1.51 | 3.53 | 1.76 | 1.52 | 27.1 | 22.7 | mihomo | 4064 | 1.0.14 |
| OneXray | 12.7 | 1.15 | 1.23 | 1.13 | 0.976 | 1.3 | 1.59 | 20 | 23.2 | Xray | 1500 | 26.10.3 |
| Tunna | 5.53 | 0.78 | 0.448 | 0.469 | 1.41 | 1.25 | 4.4 | 31.4 | 18.3 | Xray | 4064 | 1.6.7 |
| susi connect | 4.12 | 0.722 | 0.432 | 0.452 | 1.33 | 1.05 | 6.48 | 29.8 | 21.4 | Xray | 1360 | 1.12.1 |
| V2Box | 3.57 | 0.631 | 0.401 | 0.324 | 1.12 | 1.09 | 7.52 | 30.1 | 28.3 | Xray | 1500 | 10.1.9 |
| Clash Plus | 2.68 | 0.563 | 0.44 | 0.285 | 0.969 | 0.827 | 10.8 | 36.1 | 16.4 | mihomo | 1500 | 1.4.1 |
| Streisand | 2.3 | 0.502 | 0.245 | 0.245 | 1.07 | 1.01 | 12.8 | 35.7 | 21 | sing-box and Xray | 1500 | 1.6.76 |
| incy | 2.26 | 0.456 | 0.241 | 0.237 | 1.1 | 0.687 | 12.6 | 34.9 | 16 | Xray | 1500 | 2.6.4 |
| Happ | 1.93 | 0.382 | 0.224 | 0.203 | 0.747 | 0.635 | 14.6 | 43.6 | 18.2 | Xray | 1500 | 6.1.0 |
Medians over all rounds. Open an app for the range of each measure.
iPhone lab: 7 apps, 3 rounds
The same test on an iPhone: the server runs inside a test app on the phone, so only the VPN app is in the way. Ranked by speed (30%) and energy efficiency (70%).
Scroll the table sideways for every measure.
| Speed, Gbit/s | At 200 Mbit/s | At idle | Build | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| App | Scoreof 100 | Average | Upload1 connection | Download1 connection | Upload4 connections | Download4 connections | Power addedmW | Extra energyJ per GB | CPUs per GB | MemoryMB | Power addedmW | Wakeupsper s | Version |
| Egern | 91.8 | 2.46 | 2.14 | 2.7 | 2.21 | 2.82 | 155 | 6.2 | 6.87 | 6.33 | 95.6 | 4 | 2.20.0 |
| Anywhere | 63 | 3.27 | 3.18 | 2.43 | 3.54 | 3.97 | 300 | 12 | 18.7 | 6.19 | 212 | 1.3 | 2.4 |
| Shadowrocket | 49.5 | 1.16 | 0.797 | 1.42 | 0.876 | 1.82 | 271 | 10.8 | 15.4 | 6.14 | 196 | 4.25 | 2.2.92 |
| sing-box (SFI) | 25 | 0.5 | 0.287 | 0.298 | 0.936 | 0.781 | 502 | 20.1 | 30.3 | 14.9 | 17.2 | 92.2 | 1.14.7 |
| Hiddify | 16.8 | 0.342 | 0.265 | 0.238 | 0.509 | 0.425 | 756 | 30.2 | 52.1 | 25.9 | 88.8 | 118 | 4.0.0 |
| V2Box | 13.2 | 0.355 | 0.181 | 0.182 | 0.781 | 0.631 | 979 | 43.1 | 103 | 15.2 | 39.4 | 70.6 | 10.1.9 |
| Happ | 11.5 | 0.238 | 0.155 | 0.132 | 0.39 | 0.402 | 722 | 44.3 | 86.6 | 13.1 | 46.2 | 57.5 | 6.1.0 |
Medians over all rounds. Open an app for the range of each measure.
Energy in real use, at 200 Mbit/s
One protocol
| App | Setup | Test | mW | Times Egern |
|---|---|---|---|---|
| Egern | Reality one hop | real path | 7.6 | 1.0× |
| Anywhere | Reality one hop | real path | 23.7 | 3.1× |
| Shadowrocket | Reality one hop | real path | 33.7 | 4.4× |
| Aestate | Reality one hop | real path | 47.4 | 6.2× |
| WireGuard app 1.0.16 | WireGuard | lab | 170 | 22× |
| V2Box | Shadowsocks | lab | 188 | 25× |
| Happ | Shadowsocks | lab | 365 | 48× |
Chain + rules
| App | Setup | Test | One hop, mW | Extra, mW | Total, mW | Times one hop |
|---|---|---|---|---|---|---|
| Egern | Reality chain | real path | 7.6 | 15.7 | 23.3 | 3.1× |
| Egern | Hysteria2 chain | lab | – | – | 78 | – |
| Shadowrocket | Reality chain | real path | 33.7 | 66.7 | 100.4 | 3.0× |
| Anywhere | Reality chain | real path | 23.7 | 79.9 | 103.6 | 4.4× |
Notes on these numbers
The app's own energy, as macOS estimates it for the app's processes, at a 200 Mbit/s download (J per GB × 25). Real path = through real servers over the internet; the 14–15 Oct runs were at 100 Mbit/s, assuming the same energy per GB. Lab = a test server on the same Mac.
- Lab tests used other protocols, so they compare only roughly with the real-path results.
- Aestate was measured on 27 Oct at about 185 Mbit/s; Egern read 9 mW in that same run, so Aestate used about 5.4× Egern's energy on the day.
- Chain: the outer protocol to a first server, Shadowsocks inside it to a second, plus everyday routing rules.
- Egern's Hysteria2 chain is from a lab test on 25 Oct; Hysteria2 alone was not measured, so it has no split. In that lab Egern's Reality chain read 15 mW, so a real path may read higher.
- Anywhere's chain was measured on version 2.4, its one hop on 2.3. Aestate has no domain routing, so it has no chain.
Who sees what, hop by hop
Each route from your device to the website.
Classic VPN
Censor sees a vpn handshake. Servers: 1 server. Delay: +1 RTT (WireGuard).
| Hop | Final website | Your IP | Site name | Content | What happens |
|---|---|---|---|---|---|
| Your device VPN app | Sees it | Sees it | Sees it | Sees it | The VPN wraps every IP packet (TCP, UDP, everything) in an encrypted tunnel. |
| Censor DPI | Can't see it | Sees it | Can't see it | Can't see it | Your country's filter (DPI) sits on the path. It sees every packet's addresses and timing, and tries to classify the traffic. WireGuard's first byte is a plaintext message type and its handshake has fixed sizes, so it is easy to recognise. Iran blackholed it after the handshake in June 2026; Russia blocks it since 2023. |
| VPN server decrypt + NAT | Sees it | Sees it | Sees it | Partly: only non-HTTPS | Decrypts the packet and sends it on from its own IP. It sees every site you visit (but not inside HTTPS). |
| Website sees the VPN IP | Sees it | Can't see it | Sees it | Sees it | The site sees the VPN server's address, not yours. |
REALITY self-steal
Censor sees visit to your own site. Servers: 1 server. Delay: +2 RTT (TCP + TLS).
| Hop | Final website | Your IP | Site name | Content | What happens |
|---|---|---|---|---|---|
| Your device VLESS client | Sees it | Sees it | Sees it | Sees it | Same REALITY client, but the name it shows is your own domain. |
| Censor DPI | Can't see it | Sees it | Partly: your domain | Can't see it | Your country's filter (DPI) sits on the path. It sees every packet's addresses and timing, and tries to classify the traffic. Name, DNS and IP all agree, so the name-vs-IP check finds nothing. In Russia one user reported it working without problems (Oct 2026). In this site's own field test it ran at ~70 Mbps where a borrowed name got 1–2. |
| Your server REALITY + web | Sees it | Sees it | Sees it | Partly: only non-HTTPS | Failed logins go to your own web server (nginx with a real Let's Encrypt certificate) on the same machine, so probes see a real site. |
| Website sees VPS IP | Sees it | Can't see it | Sees it | Sees it | Sees your server's address. |
QUIC (Hysteria2)
Censor sees udp / http/3. Servers: 1 server. Delay: +1 RTT, 0 later.
| Hop | Final website | Your IP | Site name | Content | What happens |
|---|---|---|---|---|---|
| Your device Hy2 client | Sees it | Sees it | Sees it | Sees it | One QUIC connection carries all streams. Login is an HTTP/3 request; Salamander scrambles every packet. |
| Censor DPI | Can't see it | Sees it | Can't see it: with Salamander | Can't see it | Your country's filter (DPI) sits on the path. It sees every packet's addresses and timing, and tries to classify the traffic. Plain QUIC shows its server name (China blocks QUIC by name since Apr 2024; a 2026 study saw Iran block all QUIC at one data-centre vantage point in Jun–Jul 2025). With Salamander it's random UDP, which works until a network drops unknown UDP. |
| Hy2 server masquerade | Sees it | Sees it | Sees it | Partly: only non-HTTPS | Without the password it behaves like an ordinary HTTP/3 web server (only with Salamander off; with it on, probes can't connect at all). Port hopping (switching ports every 30 s) defeats per-port UDP throttling. |
| Website sees server IP | Sees it | Can't see it | Sees it | Sees it | Sees the server's address. |
CDN-fronted
Censor sees https to a cdn. Servers: cdn + server. Delay: +2–3 RTT, longer path.
| Hop | Final website | Your IP | Site name | Content | What happens |
|---|---|---|---|---|---|
| Your device proxy app | Sees it | Sees it | Sees it | Sees it | Connects to a CDN edge IP (Cloudflare or similar) with your domain, using WebSocket, gRPC or XHTTP requests. |
| Censor DPI | Can't see it | Sees it | Partly: your domain | Can't see it | Your country's filter (DPI) sits on the path. It sees every packet's addresses and timing, and tries to classify the traffic. It sees HTTPS to an IP shared with millions of sites, so blocking the IP is costly. It can still block your domain by name: Iran did within ~24 h in June 2026, even with packet fragmentation (one report). |
| CDN edge Cloudflare | Can't see it | Sees it | Sees it | Partly: HTTP layer only | Ends the TLS, reads the HTTP requests (paths, headers) and forwards them to your server over a second TLS connection. It can close your account. |
| Origin server Xray | Sees it | Partly: via header | Sees it | Partly: only non-HTTPS | Your server, hidden behind the CDN. It sees the CDN's address plus a forwarded header with yours. |
| Website sees origin IP | Sees it | Can't see it | Sees it | Sees it | Sees your origin server's address. |
Relay chain
Censor sees a domestic connection. Servers: 2 servers. Delay: +1 relay hop (~2× RTT).
| Hop | Final website | Your IP | Site name | Content | What happens |
|---|---|---|---|---|---|
| Your device proxy app | Sees it | Sees it | Sees it | Sees it | Connects to a server inside your own country. Domestic sites go direct (split routing). |
| ISP filter domestic DPI | Can't see it | Sees it | Partly: relay's name | Can't see it | Your ISP sees a connection to a domestic data-centre. Domestic traffic is filtered less, though in the 2026 shutdown even Iran's national network was cut at times. |
| Relay in-country VPS | Sees it | Sees it | Sees it | Partly: only non-HTTPS | Decrypts the first leg, routes local sites directly, and sends the rest abroad over one multiplexed link. Some data-centres drop UDP abroad, so this leg is often TCP-only. |
| Border gateway international | Can't see it | Can't see it | Partly: exit's name | Can't see it | Your country's filter (DPI) sits on the path. It sees every packet's addresses and timing, and tries to classify the traffic. Here it sees a data-centre talking to a foreign server: a second REALITY session, without your IP. |
| Exit server US VPS | Sees it | Can't see it | Sees it | Partly: only non-HTTPS | Sees the relay, not you. Mux (smux) cuts the new-connection delay from ~512 to ~162 ms in a field test. |
| Website sees exit IP | Sees it | Can't see it | Sees it | Sees it | Sees the exit's address. |
DNS tunnel
Censor sees dns lookups. Servers: resolver + server. Delay: Very high.
| Hop | Final website | Your IP | Site name | Content | What happens |
|---|---|---|---|---|---|
| Your device dnstt / Slipstream | Sees it | Sees it | Sees it | Sees it | Encodes your data in DNS lookups for a domain you control. |
| Domestic resolver ISP DNS | Can't see it | Sees it | Partly: your tunnel domain | Can't see it | Your country's own DNS server looks the name up for you. It must still be able to reach abroad, which some did during Iran's 2026 blackout. |
| Censor DPI | Can't see it | Can't see it | Partly: tunnel domain | Can't see it | Your country's filter (DPI) sits on the path. It sees every packet's addresses and timing, and tries to classify the traffic. It sees the resolver doing DNS lookups, which almost no network can switch off completely. |
| Your DNS server tunnel end | Sees it | Can't see it | Sees it | Partly: only non-HTTPS | Answers the lookups with your data and forwards your traffic onward. |
| Website sees server IP | Sees it | Can't see it | Sees it | Sees it | Sees your server's address. |
What works in each country, 2026
Mostly user reports on net4people, often from one ISP and a few people. Outlined = inferred or unconfirmed.
Iran
| Technique | Status | Latest report |
|---|---|---|
| AmneziaWG | Works | Reported working, Jun 2026 (n4p #628) |
| Shadowsocks / SS2022 | Works | 'Works great', late Jul 2026 (n4p #640) |
| VLESS + REALITY | Works | Working again from mid-Jun 2026; needs IP rotation. Borrowed names throttled, self-steal fine (field test) |
| Hysteria2 | Works | All ISPs except Rightel (one user), 13 Jun 2026; fails in blackouts and where UDP is dropped |
| AnyTLS | Works | All ISPs except Rightel, Jun 2026 |
| Tor Snowflake | Works | Most-used Tor transport in Iran (2025) |
| Tor WebTunnel | Works | Bridges via Telegram (2025) |
| Psiphon | Works | Main method in late Jan 2026; failed at times |
| DNS tunnels | Works | Slow, but worked through the Feb–May 2026 blackout |
| IKEv2/IPsec | Degraded | Own field test: usable only as a server-to-server overlay (Oct 2026) |
| VLESS via CDN | Degraded | Works at times (up to 150 Mbps); domains blocked within ~24 h (one report) |
| Trojan | Degraded | Domain and IP blocked after use, Jun 2026 |
| Plain HTTP disguise (port 80) | Degraded | Some ISPs only (Iranian blog, 2025); an HTTP prefix passed a protocol allowlist for 4 weeks (study, 2023–25 data) |
| WireGuard | Blocked | Blackholed after handshake, Jun 2026 (one report, n4p #628) |
| TUIC / plain QUIC | Blocked | All QUIC blocked at one vantage point, Jun–Jul 2025 (FOCI 2026) |
| ClientHello fragmentation | Blocked | Stopped working ~27 Jul 2026: fragmented hellos dropped (n4p #640); reassembly seen from Jun (n4p #628) |
| MASQUE | Blocked | All QUIC blocked at one vantage point, Jun–Jul 2025 (n4p #648) |
| OpenVPN | No data (inferred) | No 2026 source |
| VMess | No data | Largely replaced by VLESS |
| NaiveProxy | No data | – |
| Tor obfs4 | No data | – |
China
| Technique | Status | Latest report |
|---|---|---|
| VLESS + REALITY | Works | Replay-probe risk on some targets (Xray #6121) |
| VLESS via CDN | Works (inferred) | CDN collateral damage; no 2026 measurement |
| NaiveProxy | Works (inferred) | Chrome network stack; no 2026 measurement |
| Tor WebTunnel | Works | Works where obfs4 fails |
| Shadowsocks / SS2022 | Degraded | Classic SS mostly blocked, SS2022 degraded: fully-encrypted detection since 2021 (USENIX 2023) |
| Trojan | Degraded | TLS-in-TLS detectable |
| Hysteria2 | Degraded | Residential UDP QoS cuts long sessions |
| TUIC / plain QUIC | Degraded | QUIC blocked by name since Apr 2024 |
| Tor Snowflake | Degraded | Blocked in Fujian, Aug 2026 |
| WireGuard | Blocked (inferred) | Handshake fingerprintable; no 2026 source |
| OpenVPN | Blocked (inferred) | Well known; no 2026 source |
| IKEv2/IPsec | Blocked (inferred) | Unconfirmed; no 2026 source |
| VMess | Blocked | Fully-encrypted detection (USENIX 2023) |
| Tor obfs4 | Blocked | Allowlist-style filters reject it (Tor WebTunnel blog, 2024) |
| AmneziaWG | No data | – |
| AnyTLS | No data | – |
| ClientHello fragmentation | No data | – |
| Plain HTTP disguise (port 80) | No data | – |
| MASQUE | No data | – |
| Psiphon | No data | – |
| DNS tunnels | No data | – |
Russia
| Technique | Status | Latest report |
|---|---|---|
| AmneziaWG | Works | Works, but servers caught in mass IP bans, Aug–Sep 2026 (n4p #671) |
| VLESS + REALITY | Works | Self-steal reported working (one user, Oct 2026); popular VPS providers' IPs mass-banned |
| Tor Snowflake | Works | DTLS filtering from 30 Mar 2026; one user: working again in Sep (n4p #603) |
| Shadowsocks / SS2022 | Degraded | Reportedly blockable (2024); keep first packets small (n4p #673) |
| Trojan | Degraded | ~15–20 KB freeze on TLS to foreign hosting, since mid-2025 (n4p #490) |
| Hysteria2 | Degraded | UDP blocked on mobile |
| TUIC / plain QUIC | Degraded | QUIC v1 name filter since 2023 |
| Tor obfs4 | Degraded | Blocked on some 4G networks, late 2024 |
| Tor WebTunnel | Degraded | Most bridges listed from Jun 2025 |
| WireGuard | Blocked | Blocking began 2023 (Mediazona) |
| OpenVPN | Blocked | Blocking began 2023 |
| IKEv2/IPsec | Blocked | Reportedly blocked, Jun 2024 (Moscow Times) |
| VMess | No data | – |
| VLESS via CDN | No data | Under mobile allowlists only allowlisted destinations work, whatever the protocol (n4p #579) |
| AnyTLS | No data | – |
| NaiveProxy | No data | – |
| ClientHello fragmentation | No data | – |
| Plain HTTP disguise (port 80) | No data | – |
| MASQUE | No data | – |
| Psiphon | No data | – |
| DNS tunnels | No data | National DNS may limit them |
Turkmenistan
| Technique | Status | Latest report |
|---|---|---|
| AmneziaWG | Degraded | Banned after moderate traffic (one user, Aug 2026); a low-traffic server lasted 2 years (2025) |
| VLESS via CDN | Degraded | Only thing that somewhat works; CDN IPs and domains often blocked (n4p #523, 2025) |
| WireGuard | Blocked | Likely: foreign IPs banned after moderate traffic, whatever the protocol (n4p #523) |
| OpenVPN | Blocked (inferred) | Unconfirmed |
| VLESS + REALITY | Blocked | IPs banned after moderate traffic (n4p #523) |
| Trojan | Blocked (inferred) | Likely; no 2026 source |
| Hysteria2 | Blocked | Banned with and without Salamander and port hopping |
| TUIC / plain QUIC | Blocked (inferred) | Likely; no 2026 source |
| IKEv2/IPsec | No data | – |
| Shadowsocks / SS2022 | No data | – |
| VMess | No data | – |
| AnyTLS | No data | – |
| NaiveProxy | No data | – |
| ClientHello fragmentation | No data | – |
| Plain HTTP disguise (port 80) | No data | – |
| Tor obfs4 | No data | Bridges reported blocked, Jan 2026; transport not named (Tor forum) |
| Tor Snowflake | No data | – |
| Tor WebTunnel | No data | – |
| MASQUE | No data | – |
| Psiphon | No data | – |
| DNS tunnels | No data | – |
All 58 techniques
✓ works, ~ degraded, ✕ blocked, – no data; outlined = inferred.
| Technique | Year | Iran | China | Russia | Turkmenistan | Family | Class | Connection | Phone power | What it is |
|---|---|---|---|---|---|---|---|---|---|---|
| ECH | RFC 2026 | Iran: No data | China: No data | Russia: Degraded | Turkmenistan: No data | Disguise layer | Cutting edge | TCP | Moderate | Encrypts the website name in the TLS hello. |
| Finalmask | 2026 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Disguise layer | Cutting edge | Any | High | Disguises every packet; can tunnel through DNS or ping (ICMP). |
| XDRIVE | 2026 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Transport | Cutting edge | TCP | Moderate | Moves proxy data through cloud-storage services. |
| paqet | 2026 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Tor, DNS and others | Cutting edge | Raw IP | High | Sends raw packets with KCP, skipping the normal network stack. |
| AnyTLS | 2025 | Iran: Works | China: No data | Russia: No data | Turkmenistan: No data | Looks like HTTPS | Cutting edge | TCP | Moderate | A newer TLS proxy that pads traffic to hide the TLS-inside-TLS pattern. |
| ShadowQUIC | 2025 | Iran: Blocked (inferred) | China: No data | Russia: No data | Turkmenistan: No data | QUIC / UDP | Cutting edge | QUIC | High | A new QUIC proxy that borrows a real site's handshake, like REALITY for QUIC. |
| Sudoku | 2025 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Random-looking proxy | Cutting edge | TCP | Moderate | A new protocol aimed at looking less random than Shadowsocks-style traffic. |
| TrustTunnel | 2025 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Looks like HTTPS | Cutting edge | TCP + UDP | Moderate | A VPN-style tunnel over HTTP/2 and HTTP/3. |
| VLESS Encryption | 2025 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Disguise layer | Cutting edge | Any | Moderate | Adds post-quantum encryption inside VLESS itself. |
| Slipstream / XDNS | 2024–2026 | Iran: Works | China: No data | Russia: No data | Turkmenistan: No data | Tor, DNS and others | Cutting edge | DNS | Highest | Faster DNS tunnels from the 2026 blackouts. |
| Tor WebTunnel | 2024 | Iran: Works | China: Works | Russia: Degraded | Turkmenistan: No data | Tor, DNS and others | Cutting edge | TCP | High | Hides a Tor bridge behind a real website. |
| XHTTP | 2024 | Iran: Degraded | China: Works (inferred) | Russia: No data | Turkmenistan: Degraded | Transport | Cutting edge | TCP | Moderate | Splits the proxy into ordinary HTTP requests. Passes almost any CDN or middlebox. |
| AmneziaWG | 2023 · v2 2025 · v3.1 2026 | Iran: Works | China: No data | Russia: Works | Turkmenistan: Degraded | Classic VPN | Cutting edge | UDP | Low | WireGuard with disguise: junk packets and random sizes so the handshake doesn't match. |
| HTTPUpgrade | 2023 | Iran: Degraded (inferred) | China: Works (inferred) | Russia: Works (inferred) | Turkmenistan: Blocked (inferred) | Transport | Popular | TCP | Moderate | A lighter WebSocket: does the upgrade, then sends raw bytes. |
| Hysteria2 | 2023 | Iran: Works | China: Degraded | Russia: Degraded | Turkmenistan: Blocked | QUIC / UDP | Popular | QUIC | High | A QUIC proxy built for bad, lossy links. Looks like an HTTP/3 website. |
| REALITY | 2023 | Iran: Works | China: Works | Russia: Works | Turkmenistan: Blocked | Disguise layer | Popular | TCP | Moderate | Borrows a real website's TLS handshake. No domain or certificate needed. |
| Salamander | 2023 | Iran: Works | China: Degraded | Russia: Degraded | Turkmenistan: Blocked | Disguise layer | Popular | UDP | High | Scrambles Hysteria2 packets so they no longer look like QUIC. |
| TUIC v5 | 2023 | Iran: Blocked | China: Degraded | Russia: Degraded | Turkmenistan: Blocked (inferred) | QUIC / UDP | Popular | QUIC | High | A lean QUIC proxy that can send data with zero round trips. |
| Juicity | 2023 | Iran: Blocked (inferred) | China: Degraded (inferred) | Russia: Degraded (inferred) | Turkmenistan: Blocked (inferred) | QUIC / UDP | Legacy | QUIC | High | A QUIC proxy from the dae project. Gone quiet. |
| MASQUE | RFC 2022–23 | Iran: Blocked (inferred) | China: No data | Russia: No data | Turkmenistan: No data | Classic VPN | Cutting edge | QUIC | High | A VPN carried inside HTTP/3, so it looks like ordinary web traffic. Powers Cloudflare WARP. |
| ShadowTLS v3 | 2022 · v3 2023 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Looks like HTTPS | Popular | TCP | Moderate | Does a real handshake with a trusted website, then switches to your proxy data. |
| Shadowsocks 2022 | 2022 | Iran: Works | China: Degraded | Russia: Degraded | Turkmenistan: No data | Random-looking proxy | Popular | TCP + UDP | Moderate | The modern Shadowsocks: fixed keys, replay protection, faster. |
| XTLS Vision | 2022 | Iran: Works | China: Works | Russia: Works | Turkmenistan: Blocked | Disguise layer | Popular | TCP | Moderate | Hides the 'TLS inside TLS' pattern that gives most TLS proxies away. |
| Cloudflare quick tunnel | 2021 | Iran: Degraded (inferred) | China: No data | Russia: No data | Turkmenistan: No data | Tor, DNS and others | Popular | TCP | Moderate | A free, throwaway Cloudflare hostname that reaches your server even when its IP is blocked. |
| Mieru | 2021 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Random-looking proxy | Popular | TCP + UDP | Moderate | An encrypted proxy with random padding, from an independent developer. |
| gRPC | 2021 | Iran: Degraded | China: No data | Russia: No data | Turkmenistan: No data | Transport | Popular | TCP | Moderate | Carries the proxy as gRPC calls over HTTP/2. CDN-friendly. |
| DNS tunnel (dnstt) | 2020 | Iran: Works | China: No data | Russia: No data | Turkmenistan: No data | Tor, DNS and others | Popular | DNS | Highest | Hides data inside DNS lookups. Very slow, but works when almost nothing else does. |
| VLESS | 2020 | Iran: Works | China: Works | Russia: Works | Turkmenistan: Blocked | Looks like HTTPS | Popular | TCP | Moderate | Xray's flagship: a minimal protocol that leaves encryption to TLS or REALITY. |
| Hysteria (v1) | 2020 | Iran: Blocked (inferred) | China: No data | Russia: No data | Turkmenistan: No data | QUIC / UDP | Legacy | UDP | High | The first Hysteria. Replaced by Hysteria2. |
| NaiveProxy | 2019 | Iran: No data | China: Works (inferred) | Russia: No data | Turkmenistan: No data | Looks like HTTPS | Popular | TCP | Moderate | Uses Chrome's own network code, so its fingerprint is a real browser's. |
| Tor Snowflake | ~2019 | Iran: Works | China: Degraded | Russia: Works | Turkmenistan: No data | Tor, DNS and others | Popular | UDP | High | Reaches Tor through volunteers' browsers, looking like a video call. |
| WARP / multi-exit outbound | 2019 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Tor, DNS and others | Popular | UDP | Moderate | Sends the server's outgoing traffic through Cloudflare WARP (or Psiphon, Proton) to fix sites that block data-centre IPs. |
| Cloak | 2018 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Disguise layer | Popular | TCP | Moderate | Makes OpenVPN or Shadowsocks look like HTTPS. |
| MTProto proxy | 2018 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Tor, DNS and others | Popular | TCP | Moderate | Telegram's own proxy. Works for Telegram only. |
| Snell | ~2018 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Looks like HTTPS | Popular | TCP | Moderate | Surge's own closed-source protocol. |
| HTTP/2 transport | ~2018 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Transport | Legacy | TCP | Moderate | The old HTTP/2 transport. Removed from Xray in Dec 2024 (merged into XHTTP). |
| Brook | 2017 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Random-looking proxy | Popular | TCP + UDP | Moderate | A one-binary proxy with its own simple protocol and a WebSocket mode for CDNs. |
| ClientHello fragmentation | ~2017 · Xray 2023 | Iran: Blocked | China: No data | Russia: No data | Turkmenistan: No data | Disguise layer | Popular | TCP | Moderate | Cuts the TLS hello into pieces so a simple filter can't read the site name. |
| Trojan | 2017 | Iran: Degraded | China: Degraded | Russia: Degraded | Turkmenistan: Blocked (inferred) | Looks like HTTPS | Popular | TCP | Moderate | Looks exactly like a normal HTTPS website, because it is one. |
| uTLS fingerprints | ~2017 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Disguise layer | Popular | TCP | Moderate | Makes the proxy's TLS hello look exactly like Chrome, Firefox or Safari. |
| WebSocket | 2016 | Iran: Degraded | China: Works (inferred) | Russia: No data | Turkmenistan: Degraded | Transport | Popular | TCP | Moderate | Carries the proxy inside a WebSocket, so it can pass through CDNs like Cloudflare. |
| HTTP header disguise | ~2016 | Iran: Degraded | China: No data | Russia: No data | Turkmenistan: No data | Disguise layer | Legacy | TCP | Moderate | Dresses the connection up as plain, unencrypted web traffic, e.g. a request to google.com. |
| mKCP | 2016 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Transport | Legacy | UDP | High | An old UDP transport that trades bandwidth for speed on lossy links. |
| simple-obfs / v2ray-plugin | 2016–2019 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Disguise layer | Legacy | TCP | Moderate | Old Shadowsocks plugins that wrap it in fake HTTP or WebSocket. |
| WireGuard | 2015 · stable 2020 | Iran: Blocked | China: Blocked (inferred) | Russia: Blocked | Turkmenistan: Blocked | Classic VPN | Popular | UDP | Low | The modern VPN: small, fast, and silent when idle. Not built to hide. |
| ShadowsocksR | 2015 | Iran: No data | China: Blocked (inferred) | Russia: No data | Turkmenistan: No data | Random-looking proxy | Legacy | TCP + UDP | Moderate | A Shadowsocks fork with obfuscation plugins. Unmaintained. |
| VMess | 2015 | Iran: No data | China: Blocked | Russia: No data | Turkmenistan: No data | Random-looking proxy | Legacy | TCP | Moderate | V2Ray's original protocol. Still everywhere, but replaced by VLESS. |
| Tor obfs4 | 2014 | Iran: No data | China: Blocked | Russia: Degraded | Turkmenistan: No data | Tor, DNS and others | Popular | TCP | High | Tor's default bridge disguise: random-looking bytes. |
| Tor meek | 2014 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Tor, DNS and others | Legacy | TCP | High | Domain fronting through a CDN. Mostly dead since 2018. |
| Tailscale / ZeroTier | 2013–2019 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Classic VPN | Popular | UDP | Moderate | Mesh VPNs that link your own devices directly, punching through home routers. |
| Shadowsocks | 2012 · AEAD 2017 | Iran: Works | China: Blocked | Russia: Degraded | Turkmenistan: No data | Random-looking proxy | Popular | TCP + UDP | Moderate | The original anti-censorship proxy: encrypt everything so it looks like random noise. |
| SSTP | 2007 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Classic VPN | Legacy | TCP | Highest | Microsoft's VPN over HTTPS. Windows-only and slow. |
| IKEv2/IPsec | 2005 | Iran: Degraded | China: Blocked (inferred) | Russia: Blocked | Turkmenistan: No data | Classic VPN | Popular | UDP | Moderate | The VPN built into iPhones, Macs and Windows. Fast and runs in the system kernel. |
| OpenVPN | 2001 | Iran: No data (inferred) | China: Blocked (inferred) | Russia: Blocked | Turkmenistan: Blocked (inferred) | Classic VPN | Popular | TCP + UDP | Highest | The classic open-source VPN. Flexible, but heavy and easy to detect. |
| L2TP/IPsec | 1999 | Iran: Blocked (inferred) | China: Blocked (inferred) | Russia: Blocked | Turkmenistan: No data | Classic VPN | Legacy | UDP | Moderate | A tunnel (L2TP) wrapped in IPsec for encryption. Double-wrapped and slow. |
| SOCKS5 / HTTP proxy | 1996 | Iran: Blocked (inferred) | China: Blocked (inferred) | Russia: Blocked (inferred) | Turkmenistan: Blocked (inferred) | Random-looking proxy | Popular | TCP + UDP | Moderate | The basic, unencrypted proxy. Used locally between an app and its core. |
| PPTP | ~1996 | Iran: Blocked (inferred) | China: Blocked (inferred) | Russia: Blocked (inferred) | Turkmenistan: Blocked (inferred) | Classic VPN | Legacy | TCP | Low | Microsoft's early VPN. Broken encryption; removed from iPhones. |
| SSH tunnel | 1995 | Iran: No data | China: No data | Russia: No data | Turkmenistan: No data | Tor, DNS and others | Popular | TCP | Moderate | Uses a normal SSH login as a proxy. Everywhere, but slow over distance. |