Mayak Blog

Telegram not working — even with a VPN: why, and what to do

Feeds load, videos play, sites open — but Telegram is not working: it hangs on “Connecting…”, and messages sit with a little clock and never arrive. You switch on a VPN — and nothing changes, Telegram stays silent with it too. A friend on the same carrier has no trouble, and you do. Below — why “the internet works but one app is silent” is built differently from “the network is down”, how to tell your case apart in a couple of minutes, what a tunnel changes here and what it does not — and our measurement from four machines in four countries, together with what it does not say.

In short. If the internet works but Telegram does not, the cause is almost always not in the app but on the road to its servers. The sign is simple: other sites open, echo packets reach Telegram’s addresses, yet a connection to them is never established — and the app says “Connecting…” forever. We measured this from our own machines: from the one in Russia, not a single connection out of 21 attempts; from the three abroad, all answered within fractions of a second. A tunnel changes the address the request leaves from, but sometimes it is up and the app is still silent — we go through four reasons separately.

We make Mayak — a private network with an app for Android. Everything said about us below is said with measurements: how our servers see the network, what the tunnel changes, and what those measurements do not say. You can check it on your own network for free. 3 days free after you confirm your email, no card needed.

Why Telegram does not work when the internet does

“Not working” for an app and “not working” for the network look the same but are built differently. Let us go step by step through what actually happens when you open the messenger.

First the phone looks up the address, then it knocks on it. These are two different actions, and they break separately. Looking up the address is quick and inconspicuous; establishing a connection to that address is a separate step, and that is the one that most often fails for people. From the outside it looks like “the app froze”: it is waiting for an answer that does not come, and honestly says “Connecting…” for an hour if need be.

An echo is not proof. The most common mistake when checking on your own: “I ping it, the ping goes through, so the connection is there.” An echo packet and a connection are different things and travel by different rules. The road can be alive for an echo and still break off for a connection — that is exactly what we saw in our measurement, and the numbers are below.

The messenger has several addresses, and they serve different purposes. The website describing it is one thing; the servers the messages themselves travel through are quite another, and they live at other addresses. So “Telegram’s page opens in my browser but the app does not work” is not a contradiction but normal: these are two different roads. The reverse happens too.

Every network has its own road. Your path to these addresses and your friend’s path are two different routes, even if both phones show the same carrier. Hence the most galling part: it works for them and not for me, and both are telling the truth. Where exactly your roads part cannot be seen from the phone — only the result can.

The conclusion this section was written for: if everything else on the network works and one app is silent, reinstalling it is usually pointless. It is not the app that breaks but the road to it.

How to tell your case apart in a couple of minutes

The check has one purpose: to separate “something is wrong in my phone” from “something is wrong on the way to these servers”. You need no terminal, only the phone itself. Go in order and note the answers.

  1. Open something else on the same network. Any site, any other feed. If it opens, you have internet, and the problem is not “the internet is gone”.
  2. Look at what the app itself says. An endless “Connecting…” is not a freeze, it is an honest report: the connection is not being established. A message with a little clock that never goes out means the same.
  3. Switch networks. From Wi-Fi to mobile data and back, within the same minute. A different result on the two networks means the phone and the app have nothing to do with it — a particular network is to blame.
  4. Try a second device on the same network. If it is silent there too, it is definitely not the phone.
  5. Turn on the tunnel and restart the app. Restart it, do not just switch back to it: the app may have been holding on to an old failed attempt. If it works after that, the bottleneck was on the road; if not — the four reasons are covered below.
What you seeWhat it means
Sites open, Telegram does not work You have internet. The bottleneck is on the road to these particular servers, not in the phone
An endless “Connecting…” The connection is not being established. Reinstalling the app does not change that
Works on Wi-Fi, not on mobile data (or the other way round) A particular network is to blame, not the device and not the service
Silent on two devices at once It is not the phone. Look at the network and the road
Text goes through, but pictures and videos do not load Looks like packet size — a separate problem with its own fix
With the tunnel on, nothing at all goes through That is no longer about one app — “connected but no internet” is covered separately
The tunnel does not come up at all Then start with it: why it will not connect, and if all of them are silent at once — when no VPN connects

Our measurement: where Telegram answers from, and where it does not

We do not retell the news — we measure what we can reach ourselves. On 22 September 2026, between 15:25 and 15:35 Moscow time, from each of our machines we tried to establish a connection with five Telegram addresses — and in the same minute, from the same machine, with control addresses that have nothing to do with Telegram.

Measured fromFive Telegram addressesControl addresses in the same minute
Our machine in Russia The connection was never established All opened
The Netherlands All five, 0.01–0.20 s All opened
Poland All five, 0.04–0.20 s All opened
Germany All five, 0.02–0.17 s All opened
A work machine in Moscow that reaches the internet through Kazakhstan Four of five, 0.19–0.30 s; the fifth did not open All opened

We rechecked the Russian row three times. Three rounds with a pause, now against seven addresses: 21 attempts out of 21 — the connection was never established. In the same minute, from the same machine, Cloudflare, GitHub, Yandex and our own core server opened — so the machine’s network was alive.

And the echo was getting through meanwhile — 43–195 ms to those very addresses. Here it is, the case from the first section: the route is alive, but the connection is not established. A hop-by-hop trace breaks off after the ninth hop, inside the provider’s own network. The machine has no rules of its own about these addresses — we checked that too, so as not to mistake our own setting for someone else’s.

And it is about the address, not the port. From the same machine we knocked on three addresses, each on three different ports — the ones sites and messengers usually use. Not one of the nine opened. For a person this is an important difference: not “the site is not answering” and not “one service is acting up”, but the whole direction does not get through. That is why advice like “change the port” does not help: the port has nothing to do with it.

⚠️ What this measurement does NOT say. Our machine in Russia sits in a datacentre of one hosting company, and that is exactly one address from one network. It does not follow that it is “the same for everyone across Russia”: mobile carriers and home internet providers each have their own picture, and it changes by the hour. We name no shares and no percentages “for the country” — we have nothing to measure them with.

What the tunnel changes here. In the same ten minutes we brought up a real tunnel to the Netherlands line — not “on paper” but with an instrument: the configuration was issued, the handshake went through in a couple of seconds, the packet size was 1397 bytes, and the exit address matched the address of the Netherlands server. So the request goes out to the network from there — and from there, per the row above, Telegram answers within hundredths of a second.

This has to be read strictly, and we will say plainly how it is. We did not measure the end-to-end path “phone → tunnel → Telegram” as a whole. We measured two links separately: traffic leaves from the server’s address, and Telegram answers that address. A chain of two measured links and a measured chain are not the same thing, and we will not pass one off as the other.

Check it on your own network. Our measurement speaks about our machines in datacentres — it knows nothing about your network, and there is one way to find out: try it on your own carrier, on your own Wi-Fi, on your own evening. The account is created right in the app, and we ask for no payment details for the trial. A confirmed email opens the access — 3 free days.

The tunnel is on, and Telegram still does not work

The most galling case, and it has four explanations. They are different and are checked one at a time; three of the four you can check yourself without leaving the phone.

1. The app has been routed around the tunnel. Apps of this kind have split tunneling: the selected apps do not go through the tunnel but the way they did before. If the messenger ended up on that list, the tunnel is on but nothing has changed for it, and the icon in the notification shade is misleading. In our app the setting is called “Split tunneling”, and the app’s main screen has a preset bar with a ready-made set, “RU Apps” — Russian services that a tunnel only gets in the way of (how it is turned on). If the app sees that you go online from Russia, it turns this set on by itself once and tells you so directly. Telegram is not in it — we checked this on 22 September 2026 against the file itself in the production build, not from memory — so in our app it goes through the tunnel by default. It is worth looking through the whole list: the set is ours, but the ticks in it are yours.

2. Packet size. A very recognisable sign: text goes through, but pictures, voice messages and videos do not load or stop halfway. Small packets get through, large ones do not. This is not about the messenger at all, and it is not fixed inside it: how it works and what to adjust — in a separate write-up. In our measurement the tunnel came up with a packet size of 1397 bytes — a chosen value, not a random one, and it is chosen precisely for cases like this.

3. The network does not let everything through, and the tunnel is not among what it lets through either. It happens that on someone’s mobile data a small set of services opens and the rest does not, and the connection to us does not come up either. The sign: neither the messenger nor the VPN works, yet a couple of familiar pages open just fine. This is a separate mechanism, and we have covered it in detail — why only part of the internet opens. There, too, is the answer to the question we are asked most often in this combination: “my friend has the same carrier, and everything works for them”. A carrier is not one network for everyone: the city, the district and even the particular point of connection give different roads. Our measurements by carrier live not here but in the write-up on mobile internet, and we will not retell them here.

4. The tunnel did not actually come up. The icon is there but there is no connection — that happens too, and then all the reasoning above is beside the point. It takes a minute to check: four step-by-step checks — has the address changed, is traffic flowing. If it has not changed and is not flowing, go on here: the tunnel will not connect, it worked and stopped, the app will not start. And if all of them are silent at once, that is when no VPN connects, and that case is covered there in full, together with what is happening with the connection today.

Why we have anything to say about this case at all: the app does not travel by one road. If the first one fails, it tries the next by itself, and the person does not have to choose anything by hand — how the connection steps work is covered separately.

What we do not promise

In plain words, so that nothing extra can be read into the sections above.

  • That a particular service will work for you. We make no such promises about any service or any app. What you get depends on your network, your hour and your route — and those we have not measured and cannot measure.
  • That our numbers are a picture of the country. One machine in one datacentre is one address and one route. We draw no shares, no percentages and no “what works where” maps.
  • That tomorrow will be like yesterday. The measurement was taken on one day, within ten minutes. The network changes faster than articles come out, and we do not rewrite this page every evening.
  • That we answer for carriers and home internet providers. The first half of the road — from your device to us — remains entirely yours, and no tunnel remakes it.

What we really can do is answer for our own stretch of the way, measure it and show it as it is, together with what the measurement does not say. And let you check for free, because on your network the answer will be your own anyway.

What we tested

  • Where Telegram answers from — by connection, not by echo. 22 September 2026, 15:25–15:35 Moscow time, from four of our machines (Russia, the Netherlands, Poland, Germany) and from a work machine in Moscow that reaches the internet through Kazakhstan. Five addresses, plus non-Telegram control addresses from the same machine in the same minute — without them a “did not open” row would read as “the machine was down”.
  • The Russian row — three times, against seven addresses. Three rounds with a pause, 21 attempts, not one connection established. That rules out a one-off miss: one failure can be chance, twenty-one cannot.
  • That it is not our own setting. The filtering rules on that machine were checked: not a single mention of these networks, and outgoing traffic is allowed by default. Without this check we could have mistaken our own setting for someone else’s.
  • That it is about the address, not the port. Three addresses, three different ports each — nine attempts, not one successful.
  • A live tunnel — with an instrument, not on paper. A real connection to the Netherlands line was brought up: the configuration was issued, the handshake went through, the packet size was 1397 bytes, and the exit address matched the server’s address.
  • Our ready-made “RU Apps” set — against the file in the production build of the app, not from memory: Telegram is not in it, checked on 22 September 2026.
  • We did NOT measure the “text goes through, pictures do not” sign here — it is how networks work in general, and we call it exactly that; our own packet-size measurements live in a separate write-up.

What we do not know

It is more honest to list these than to keep quiet: the measurement is a one-off, and you can draw exactly as many conclusions from it as it contains.

  • We did not measure the end-to-end path “phone → tunnel → Telegram”. Neither from a phone nor over a mobile carrier. We have two measured links, not a measured chain, and we call it exactly that.
  • How things are with carriers and home internet providers. Our machines sit in datacentres. We do not answer for other people’s networks and do not paint a portrait of the country from one address.
  • What tomorrow brings. Roads change without warning, sometimes within a day. Our measurement is ten minutes on 22 September 2026, and nothing more.
  • Why exactly it is so. We see where the path breaks off — and do not see whose decision that is or for how long. We do not make up explanations.
  • How long it has been going on for you. Our connection log does not store which sites and services you visit — so we have nothing to answer this question with.

Short answers

Why does Telegram not work when the internet does? Because “the internet works” is about the road in general, while the messenger needs a road to particular addresses. That road can break off separately from everything else: other sites open, the echo gets through, but the connection is not established.

Telegram is not working even with a VPN — what do I do? Check four things in turn: whether split tunneling has routed the app around the tunnel, whether packet size is the problem (the sign: text goes through, pictures do not), whether this is the case where the network does not let the tunnel itself through, and whether the tunnel came up at all. All four are covered above.

Will a VPN help if Telegram is not working? We will not promise that about any service. The mechanics are these: the tunnel changes the address the request leaves from — and if the problem was the road, another road may turn out better. If the problem is in your stretch — your Wi-Fi, your carrier’s network, the device itself — the tunnel does not change that.

Will reinstalling the app help? If one app is silent and the rest of the network works — almost certainly not. It is not the app that breaks but the road to it; reinstalling does not touch the road.

My friend has the same carrier and everything works for them — why? Because “the same carrier” does not mean “the same network”: the city, the district and the point of connection give different routes. It is easy to check — by switching networks within the same minute.

Text goes through but pictures and voice messages do not load — is that the same thing? No, that is a separate problem: small packets get through, large ones do not. We cover it separately, and it is fixed separately too.

And if nothing at all goes through with the tunnel on? Then it is not about the messenger but about the connection: “connected but no internet” is broken down cause by cause. And if no connection comes up at all — start here.

Does your measurement mean it is like this for everyone in Russia? No, and we will repeat it on purpose: one address was measured, in a datacentre of one hosting company. There is not a single share or percentage “for the country” in this article, because we have nothing to measure them with.

Try Mayak

A secure connection for Android and measurements we publish as they are — together with what they do not say. 3 days free once you confirm your email, no card needed.

The account is created right in the app. A confirmed email opens the access — 3 free days.