Russia VPN Bypass State โ April 2026
- title
- Russia VPN Bypass State โ April 2026
- type
- analysis
- summary
- Time-stamped snapshot of which circumvention methods currently work in Russia, tiered by reliability
- tags
- russia, censorship, vpn, roskomnadzor, snapshot
- sources
- rkn-current-vpn-measures
- created
- 2026-04-21
- updated
- 2026-05-06
Time-bound note. This is a snapshot from late April 2026, synthesized from ntc.party community discussions and public statements by circumvention tool maintainers (Amnezia, Xray). The blocking landscape changes on a weekly basis โ if you're reading this and the dates in created / updated are more than two or three months old, verify against a current source (ntc.party) before acting on it.
The headline finding: protocol choice matters less than where your exit IP lives. The dominant filter in 2026 is TSPU's CIDR whitelist applied to mobile carriers, with SNI filtering and VLESS-Reality statistical fingerprinting layered on top. A poorly-configured VLESS+Reality on a Hetzner IP dies faster than a well-configured AmneziaWG 2.0 on a whitelisted Russian CIDR.
The three filters to beat
- CIDR whitelist โ applied at IP-routing layer, mostly on mobile carriers (Megafon, Beeline, MTS, T2, Yota). Connections to non-whitelisted subnets (Hetzner, DigitalOcean, OVH, Cloudflare ranges) get dropped or throttled. Bypass: endpoint IP inside a whitelisted Russian-hoster subnet. See whitelist-internet-blocking.
- SNI whitelist โ DPI inspects the TLS ClientHello's SNI and drops unknown hostnames. Bypass: outer SNI is a whitelisted domain (
max.ru,sun6-21.userapi.com, etc.) โ Reality, ShadowTLS, and domain-fronting-style tricks all attack this layer. - Statistical fingerprinting โ traffic-pattern analysis of TLS handshakes, first-packet entropy, flow timing. Catches plain VLESS-Reality-over-TCP, old Shadowsocks, plain WireGuard. Bypass: transport layers that look like ordinary web traffic (XHTTP, HTTP/2, HTTP/3) and padding/mux on raw streams.
VLESS-based methods
Working:
- VLESS + Reality + XHTTP โ the new default since late 2025. XHTTP looks like ordinary HTTP/2 or HTTP/3 traffic, so it survives the statistical fingerprinting that now catches plain Reality-over-TCP in some regions.
- VLESS + Reality + XHTTP + "steal oneself" โ serve your own real domain with its own cert at 443, fall back to nginx for non-proxy traffic. Survives active probing because the server is the claimed site.
- VLESS on whitelisted SNI + whitelisted CIDR โ the vpn-configs-for-russia subscription files ship exactly this combination. The hard part is the CIDR: personal VPSes on OVH/Hetzner/DO get dropped even with a perfect SNI.
- Cascade through RU VPS โ foreign VPS โ first hop on a whitelisted Russian subnet, second hop is the real foreign exit. Throughput hit is large (~80% reported, 500โ90 Mbps), but it sidesteps the CIDR filter.
Degraded / dropping:
- VLESS + Reality + Vision over plain TCP on non-whitelisted CIDRs โ intermittent drops across mobile operators in Povolzhye, St. Petersburg, Yekaterinburg.
- VLESS + gRPC + nginx โ reported full collapse on some connections April 10-11 2026, requiring server reboots for brief recovery.
- WebSocket without Reality โ detectable via statistical analysis; only useful behind a CDN.
Non-VLESS methods
Tier 1 โ Actively working
- AmneziaWG 2.0 โ dominant in the WireGuard-family slot. TSPU's February 13 2026 crackdown blocked WireGuard, OpenVPN, and old AmneziaWG; AmneziaWG 2.0 and Xray survived. The v2.0 rewrite added packet-shaping and handshake obfuscation beyond the original junk-packet design. Amnezia founder Mazay Banzaev has said RKN periodically pokes at it but hasn't broken it.
- Trojan + TLS โ works with a real cert on a non-suspicious CIDR. TLS-over-TLS fingerprint is the weakness; needs the same CIDR/SNI discipline as VLESS-Reality.
- NaiveProxy โ HTTP/2-over-TLS disguised as Chrome traffic, using real Chromium for the handshake so the fingerprint genuinely matches. Niche on ntc.party but consistently recommended for users who don't want Xray.
- Hysteria2 / TUIC โ QUIC-based, look like HTTP/3. Work on home fiber, flaky on mobile because TSPU's summer 2025 crackdown on unidentified UDP hit QUIC hard. Port 443/UDP mandatory, obfs padding essential.
Tier 2 โ Situational
- Shadowsocks (SS-2022) โ old Shadowsocks is dead from RF since summer 2025 (personal SS servers and paid SS VPNs failing simultaneously, with reachable HTTP/SSH on the same host). SS-2022 slightly better but first-packet entropy is still recognizable. Usable only when tunneled inside something else (SS-over-WebSocket-over-TLS, SS-over-v2ray-plugin with a real cert).
- Outline โ same situation, since Outline is Shadowsocks.
- Cloak / ShadowTLS v3 / ShadowQUIC โ use a legitimate website's TLS cert during the initial handshake. Work on the right CIDR with a whitelisted SNI, subject to the same CIDR-whitelist ceiling.
Tier 3 โ Dead or dying
- Plain WireGuard โ February 16 2026, vdsina hosting saw a WG block rollout based on the initial handshake. The punishment escalated: previously only the UDP port receiving the handshake got blocked, now all UDP ports above 1000 on the endpoint get blocked. UDP/53 and low ports still work on some operators as a janky workaround.
- OpenVPN โ fully blocked everywhere that matters; handshake fingerprint is textbook DPI material.
- VMess โ technically "still works" but community has abandoned it; known statistical fingerprints, nothing it offers over VLESS+Reality.
- MTProto / FakeTLS (Telegram) โ multiple ntc.party reports of midday Moscow time failures on April 1 2026. Inconsistent since, heavily operator-dependent.
- L2TP / IKEv2 / PPTP โ trivially DPI-detectable. Some Russian hosting docs still recommend them as fallback; ignore that advice.
Orthogonal tools (not tunnels, but complementary)
- zapret / GoodbyeDPI / ByeDPI / gecit โ client-side fragmentation and TTL manipulation (the fake-ClientHello desync family). Don't give you a foreign exit; just help you reach blocked resources directly by splitting or desyncing the TLS ClientHello so TSPU's SNI matcher loses the pattern. zapret/GoodbyeDPI/ByeDPI use NFQUEUE/WinDivert; gecit uses eBPF sock_ops on Linux for an in-kernel implementation. Active development, effective for YouTube and similar on many operators. Standard companion to any real tunnel.
- Cascade / multihop through RU VPS โ pattern-agnostic. Russian VPS on a whitelisted network as first hop, real foreign exit as second hop. Works with any tunnel protocol.
Hosting selection
In 2023-2024, you picked a hoster on price and latency. In 2026 the two questions that dominate everything else are: does this hoster's CIDR survive the TSPU whitelist, and does the hoster cooperate with RKN takedown notices. Everything else is secondary.
Tier 1 โ Community-recommended
- PQ.Hosting (Moldova-based) โ explicitly circumvention-friendly, 25+ country menu, doesn't respond to RKN requests. The default pick on ntc.party; because it's popular among RU users, its subnets do get poked at, but support rotates quickly when a block lands.
- 4vps.su โ small, RU-speaking support, VPN-specific reputation. Referenced in AntiZapret-VPN's deployment docs as a tested option.
- justhost.asia โ cheap, good for test servers and second exits. Shared IPs on low-tier plans are often pre-burned; dedicated-IP upgrades are clean.
- ishosting โ 40+ country footprint, VPN-tolerant ToS, not the cheapest but stable. Good for unusual geographies (SE Asia, LatAm).
- VDSina โ usable with protocol caveat. On February 16 2026 a WireGuard block rolled out on VDSina's Netherlands hosting based on the initial handshake, meaning the CIDR is known-targeted. Xray-stack still works there; WireGuard family doesn't.
- waicore, senko.digital, hshp.host โ smaller, less-scrutinized. The unpopularity is the advantage: lower chance of their CIDRs being on TSPU's "suspicious" list, traded against lower SLA expectations.
- u1host โ publishes current block-status docs (their February 2026 VPN troubleshooting page is one of the better real-world references). Worth choosing just for the support quality.
Tier 2 โ Use with caution
- Aeza (aeza.net / aeza.ru) โ on December 3 2025, Aeza began sending clients notifications about possible service termination for VPN use, reportedly tied to an RKN-provided list covering the
138.124/16subnet. Still used on recent ntc.party setups (e.g. "aeza + Xray/Reality + masquerade as max.ru"), but with the understanding that Aeza will terminate or swap your IP under pressure. Acceptable for personal use on non-RU-resident IPs; don't rely on their Russian-address plans. - Timeweb Cloud, Selectel, REG.RU, Beget โ Russian companies that comply with RKN as a matter of corporate survival. Useful only as the RU-side first hop in a cascade, never as your foreign exit.
- VK Cloud โ increasingly referenced specifically because it's inside the mobile operator whitelists โ you wouldn't use it alone, but as the RU endpoint in a cascade it punches through the CIDR filter where nothing else does. See the GitHub "russia-whitelist" discussion for the VK-Cloud-as-first-hop pattern.
Tier 3 โ Avoid for new deployments
- Hetzner, OVH, DigitalOcean, Vultr, Contabo โ their entire CIDR ranges are now TSPU-flagged. Not because the hosters are bad (they're the best on price/performance), but because reaching them from Russian mobile networks requires an intermediate node with a whitelisted IP. Home-fiber users still sometimes get through; mobile users increasingly don't.
- Cloudflare (as origin) โ widely reported as throttled or dropped on Russian networks. Useful as an intermediate CDN in a tunnel, not as your direct endpoint.
Selection matrix
- Personal exit: PQ.Hosting or 4vps.su in a less-mainstream country โ Moldova, Serbia, Finland all currently beat DE/NL.
- Cascaded setup: RU-side on VK Cloud, Selectel, or Timeweb (small plan, cheap โ it just relays); foreign exit on PQ.Hosting, ishosting, or similar.
- Budget floor: justhost.asia or waicore; expect to rotate IPs often.
- Maximum stability: ishosting or u1host with a dedicated IP in a less-targeted country (Kazakhstan, Armenia, Serbia, Turkey).
- IPv6-only: underused but effective. IPv6 CIDRs are filtered far less aggressively than IPv4; Hetzner/OVH IPv6-only plans can still work where their IPv4 doesn't.
Operational posture
The community rule: treat VPS IPs as disposable. Budget for IP rotation (most hosters charge โฌ1-3 for a swap, or you destroy/recreate the VM). Keep Xray/Amnezia configs in version control so respinning takes 5 minutes. Run a health check from an RU mobile connection that pings your exit every few minutes and rotates before users notice.
One thing specifically not to do: don't bring up WireGuard/OpenVPN/Shadowsocks first on a brand-new IP. Those handshakes get caught immediately and burn the new address. Start any fresh IP with VLESS+Reality+XHTTP or AmneziaWG 2.0, which have much lower first-impression detection.
Practical recommendations
If setting up fresh in April 2026:
Primary: Xray latest + VLESS + Reality + XHTTP + steal-oneself on your own domain, IPv6-first, VPS on a CIDR that isn't in TSPU's tar pit (smaller EU/Asian providers, IPv6-only endpoints hold up better than Hetzner/OVH/DO). Client apps: v2RayTun, Hiddify, or karing for cross-platform.
Non-VLESS primary: AmneziaWG 2.0 for general use + Hysteria2 for home connections where UDP still flows. Trojan-TLS or NaiveProxy as fallback when mobile UDP gets squeezed. Layer zapret/ByeDPI on top for direct access to blocked sites without routing everything through a tunnel.
Endpoint reliability > protocol choice. If your exit IP is on a Russian-whitelisted CIDR and your SNI is on the state-approved list, a mediocre protocol works. If your IP is on OVH's most notorious subnet, even the best protocol dies. vpn-configs-for-russia is the public data source tracking which CIDR/SNI combinations currently survive.
Related
- whitelist-internet-blocking โ the concept page explaining the default-deny regime this analysis responds to
- vpn-configs-for-russia โ the live tracker of working CIDR/SNI combinations
- russia-v2ray-rules-dat โ companion geoip/geosite rule files for client-side routing
- tspu โ the Russian DPI/filtering system enforcing these blocks
- vless-subscription-format โ wire format for the VLESS subscriptions referenced throughout
- karing โ recommended client for the XHTTP-era VLESS stack
- amneziawg-architect โ generator for the AWG 1.0/1.5/2.0 obfuscation parameter sets, for the AmneziaWG half of the Tier 1 list
- domain-fronting โ the cousin technique used elsewhere (Iran); note the Russian context is different enough that classic domain fronting plays a smaller role