Hysteria vs VLESS Reality

On design alone, VLESS Reality on the RAW transport is the one to try if your network blocks UDP. Hysteria 2 is the one to try if UDP and QUIC get through, the link loses packets, and you can configure Brutal, Hysteria's congestion control. Neither recommendation comes from a test.

No independent head-to-head measurement of the two was found. That is a search result, not proof that none exists, but it does mean a single detection rate or a country-by-country winner needs a published method before you trust it. Sources checked 5 November 2026.

Fexyn sells a VLESS Reality product and does not ship Hysteria, so weigh this with that in mind.

The evidence does not show either one is reliable on Russian mobile networks or in Iran. A community report from Russian mobile networks describes flows freezing on both TCP and UDP, and reports about Reality in Iran conflict.

This page cannot tell you whether your network blocks UDP, and it cannot supply a test for that. Prepare a second option in advance, before you need it.

At a glance

	Hysteria 2	VLESS Reality
Carried on	QUIC over UDP	Modified TLS; TCP on the RAW transport, carrier not stated for XHTTP or gRPC
Blanket UDP block	Normal mode depends on UDP; mimic exists but is untested	On RAW, outside a UDP-only rule by design; untested
Packet loss	Brutal does not slow down for loss, if configured; no measurement of speed under loss found	No measurement of speed under loss found

The Hysteria spec says the protocol "MUST be implemented on top of the standard QUIC transport protocol", and QUIC runs over UDP (RFC 9000).

For Reality, only the RAW transport can be pinned down. RFC 8446 asks of the layer under TLS 1.3 only "a reliable, in-order data stream". The Xray RAW page says RAW was "Renamed from the former TCP transport layer", so Reality on RAW can be described as TCP. The Xray transport page states no underlying protocol for XHTTP or gRPC, the other two transports it allows with Reality, and the XHTTP page does not either. This page makes no carrier claim for those two.

VLESS and Reality are different layers

VLESS is the proxy protocol. Reality is a modified form of TLS, and the Xray documentation says it can be used only with the RAW, XHTTP and gRPC transports.

So "Hysteria vs VLESS" is really a comparison with one VLESS setup. Anything said here about disguise applies to Reality, not to VLESS in general.

VLESS, Reality and Hysteria 2 arrived separately. v2ray-core v4.27.0 listed VLESS PREVIEW 1.3 on 6 August 2020, though whether that was the first VLESS release is unconfirmed. Xray v1.8.0, whose notes present Reality, followed on 9 March 2023. Hysteria v2.0.0 came out on 2 September 2023.

Xray also added a Hysteria 2 inbound and transport on 12 February 2026.

How each one tries to blend in

Both projects document what a server does with a connection it rejects. That bears on active probing, where a censor connects to a suspected server to see how it answers.

Hysteria 2. The protocol spec says a server should act as a standard web server when authentication fails, and advises hosting real content or a reverse proxy. Version 2.11.0 (1 August 2026) added Chrome QUIC parroting, which imitates how Chrome's QUIC connections look, on by default. Version 2.13.0 (5 October 2026) updated it to Chrome 154. No independent test of parroting was found.

Obfuscation has a documented cost. The official docs say that with it on, the server "will no longer function as a valid HTTP/3 server". So turning on Salamander, Hysteria's obfuscation option, gives up the HTTP/3 disguise.

Reality. The REALITY README claims it removes server-side TLS fingerprint characteristics. That is the project's own claim. Rejected Client Hellos (the opening message of a TLS connection) are forwarded to the target, the separate site the server is set up to pass them to, while clients get a temporary certificate, so the two cases are handled differently. The README also warns that use without XTLS, the project's method for handling TLS traffic inside a tunnel, shows a TLS-in-TLS signature: a visible pattern of one encrypted connection inside another.

What changes when QUIC or UDP is blocked

Three different blocks matter here.

SNI-based QUIC filtering

Here the censor blocks QUIC connections by the domain name they state (the SNI). In China, Zohaib et al. (USENIX Security 2025) found this had been happening since 7 April 2024, done by decrypting Initial packets, the packets that open a QUIC connection. The full text reports Hysteria's own domain and some users' custom Hysteria domains blocked.

That is domain filtering. Zohaib et al. report Hysteria-related domains blocked, not Hysteria traffic detected. The filter targets QUIC, and it says nothing about whether a Reality connection would succeed on the same network. SNI whitelists are reported separately below.

Blanket QUIC block

The abstract page of Heitmann et al. (FOCI 2026) describes blanket QUIC blocking as Russia's earlier approach. If a block really covers all QUIC, Hysteria 2's normal mode and its Chrome parroting would depend on traffic the block drops. The port scope of the block and how Hysteria fares against it are unmeasured.

Blanket UDP block

By design, a rule that drops only UDP does not touch a TCP connection, and Hysteria 2's normal mode depends on UDP. No test of either protocol under a real UDP block was found. On the Reality side, that reasoning covers the RAW transport only.

Since version 2.12.0 (6 August 2026), Hysteria has had mimic, a mode that disguises the connection as TCP. According to the changelog, it is Linux only, requires root and a kernel module, costs throughput, and is a breaking change for clients. Those documented limits make it a narrow option. No independent test of it was found.

What drives Hysteria's behaviour under packet loss

The documented loss behaviour belongs to Brutal, Hysteria's congestion control. The official docs say Brutal "does not reduce its speed in response to packet loss". Per the spec, a client that sends a receive rate of 0 "doesn't know its own rx rate", and the server then "MUST use a congestion control algorithm (e.g., BBR, Cubic)". A Hysteria 2 setup therefore does not get this behaviour automatically. It depends on configuration.

Where Brutal is in use, it keeps its rate under loss. Whether that beats Reality on a lossy link is unmeasured.

The same behaviour has a cost. The abstract page of Wang et al. (FOCI 2025) says simple threshold classifiers can identify Hysteria and Brutal traffic because it ignores congestion signals. Whether any censor has deployed this is unconfirmed. The paper predates the 2026 parroting releases, and no test of whether parroting changes this result was found.

What public research shows, by country

None of the sources below compares the two protocols directly. Each describes one country at one time, so check the date before relying on it.

China

Zohaib et al. counted 58,207 domains blocked over QUIC between October 2024 and January 2025, 11,854 of them over QUIC only, with 180 seconds of residual blocking. Whether this describes 2026 is unconfirmed. The sources behind this page include no dated measurement of Reality in China.

Russia

The abstract page of Heitmann et al. says the TSPU, Russia's network censorship system, moved from blanket to SNI-based QUIC blocking between May 2022 and July 2023, at least nine months before China. Which ports it covers now is unconfirmed. No evidence was found that Russia fingerprints Hysteria's QUIC traffic specifically.

On mobile networks, net4people #490, from 27 June 2025, reports flows freezing after a small amount of data, plus whitelists of domain names (SNI) and IP address ranges (CIDR). An update says this applies to TCP and UDP. It is a community report, and if accurate it hits both protocols, so moving to TCP would not avoid it.

Iran

The abstract page of a FOCI 2026 paper says QUIC censorship "went into effect before and stayed in effect after" the June 2025 shutdown, with UDP instabilities afterwards.

Reports on Reality are disputed. net4people #628 reports RST floods (repeated packets that force a TCP connection closed) after the Reality handshake in June 2026. The author later says it works again (22 June), and another user reports an IP blocked after 4 days (25 June). No one there reports testing Hysteria, so the thread says nothing about which would do better.

Other countries

The sources behind this page hold no dated evidence for other countries, including the UAE, Saudi Arabia, Pakistan and Turkey, so this page gives no verdict for them.

Detection research not tied to one country

Aparecium is a proof of concept that detects Reality through the behaviour of NewSessionTicket, a message a TLS server sends after the handshake. No censor is shown using it. A later Xray fix is reported only secondhand, so it is unconfirmed.

Xue et al. (USENIX Security 2024) detected obfuscated proxies at an ISP with over one million users. That Reality was among them comes only from Lantern, and the status of the Vision flow, an XTLS mode, is unconfirmed.

Legal risk

Laws on VPN use differ by country. Check local law before using either protocol.

Gaps in the public evidence

As of 5 November 2026, no public source we could find covers any of the following:

Hysteria 2 and Reality tested on the same network at the same time.
Either protocol under a real UDP block.
The speed of either protocol, on clean or lossy links.
Chrome parroting or mimic against a real censor.
An independent detection rate for either protocol in any country.
Whether any censor uses the Wang et al. classifiers or the Aparecium method.

This page has no OONI data on these protocols, and OONI Explorer itself was not queried, so results may exist there. Claims from the FOCI papers rest on their abstract pages rather than their full text, and figures that appear only in secondary summaries are left out.

Until those gaps close, the most useful check is trying your options on your own network, before you need them. Blocks change, and the dates above show how quickly reports can reverse.

Fexyn does not publish a success rate for Fexyn Stealth either. Its own connection data is too thin to support one.

Disclosure

Fexyn's VLESS Reality product is Fexyn Stealth. Fexyn does not ship Hysteria, so it has a commercial interest in how this comparison reads.