How many people use Tor? The 2026 numbers, and what they actually count

Daily users, not total people: how Tor counts

Tor is built so that nobody can see who uses it, and that includes the Tor Project. So it doesn't count users. Its methods page says it counts "requests to the directories or bridges that clients make periodically to update their list of relays" and estimates users from those.

The arithmetic is simple. A client connected all day makes about 15 directory requests, but most aren't connected all day. So Tor assumes 10 requests per average client and divides the total by 10. In other words, each request stands for someone online for a tenth of a day, which is 2 hours and 24 minutes.

"We can't say how many distinct users there are. That is, we can't say whether the same set of users stays connected over the whole day, or whether that set leaves after a few hours and a new set of users arrives."

That has three consequences when you read any Tor user figure, including ours:

• It's a concurrency figure. "3.3 million users" means about that many online at once on an average day. The number of different people who open Tor in a month is larger by an unknown amount.
• Anything that makes directory requests counts. A server running Tor around the clock or a bot fetching the consensus looks like users. That matters for the 2025 spike below.
• Days with too little data are dropped. Tor skips any date where the reporting relays cover less than 10% of requests, or more than 110%. The direct-user series has no estimate at all for February 2026 and only 11 days for April 2026, so monthly averages around those gaps rest on less data.

Country comes from GeoIP lookups that relays run on client addresses. A user behind a VPN or a cloud server is counted where that server is, not where the person is.

Is Tor use growing or shrinking?

Over the almost four years charted, direct use has swung between about 2 million and 5 million in normal months, with one enormous spike. There's no clean growth line, and anyone who draws one is choosing their start date.

The quiet stretches tell you more than the peaks. From May 2024 to March 2025, and again from June to August 2025, direct use sat close to 2 million a month. Since March 2026 it has run between about 2.9 and 4.3 million, and September 2026 averaged 3.19 million. Measured against that 2024–25 floor, Tor use in 2026 is up. Measured against late 2023 and early 2024, when several months averaged above 5 million, it's down.

Bridge use is steadier and tells a clearer story. It averaged 117,000 to 143,000 a month through most of 2024 and the first nine months of 2025, and 210,000 to 231,000 a month from June to October 2026. Those are the people who need Tor to get past a block, and their number has grown by more than half in about a year.

What happened with the jump in late 2025

On 2 September 2025 the direct-user estimate hit 19,374,696, ten times what it had been in August. The September average was 13.3 million, and the estimate stayed around 7 million until January 2026.

Tor Metrics keeps a public list of events that explain jumps in its graphs, and it has no entry for this one. No censorship event, no bot campaign, no software change is listed for September 2025. The rise was concentrated in countries where Tor is already common: in September 2025, the United States averaged 3.6 million and Germany 2.1 million, against 648,000 and 310,000 in the most recent 30 days.

Ten million extra people did not start using Tor in a week and leave again. The more likely explanation is clients that make far more directory requests than a person would. The one relevant clue in Tor's record supports that reading. On 28 January 2026, Tor released version 0.4.8.22, which counts consensus downloads only after they complete. Before that, incomplete downloads, such as those caused by Tor being blocked, "would inflate apparent user count statistics". The next month with an estimate, March 2026, averaged 4.29 million, down from 7.15 million in January.

That's a correlation, not proof. February is missing entirely, and Tor hasn't said the fix caused the drop. For reading the numbers, though, the practical rule is clear: treat September 2025 to January 2026 as unreliable, and don't use it as a baseline for growth or decline.

Which countries use Tor the most

Tor usage by country looks completely different depending on which users you count. The direct-user list follows population, internet access and privacy habits. The bridge list follows censorship.

Sources: Tor Metrics direct-user table and bridge-user table for the same dates.

The United States in third place on the bridge list doesn't mean Americans are censored. The Snowflake team has noted that geolocation conflates some Iranian users with the US, so part of that figure is probably Iran. Read every country row the same way: GeoIP places users where their connection appears to come from, and that isn't always where they live.

Where Tor is blocked, and how the data shows it

Three kinds of evidence point at censorship, and each has its own blind spot.

Bridge share

Where Tor works directly, almost nobody bothers with bridges. When a country makes up a large share of all bridge users, direct connections there are probably failing. Russia, at 43% of the world's bridge users, is the clearest case.

OONI measurements

Volunteers run OONI Probe, which tries to reach Tor's directory authorities, relays and obfs4 bridges and, in a separate test, to start Tor over Snowflake. A failure is flagged as an anomaly. An anomaly is a sign of blocking, not proof of it.

Event notes

Tor Metrics logs shutdowns and blocking it knows about, which helps explain a sudden drop that would otherwise look like people losing interest.

Iran's low Tor-test figure looks like good news, and it isn't. Iran's filtering works differently from Russia's: it uses a protocol allowlist, and during a full shutdown no bridge can work at all. The Snowflake test, at 80% anomalies, is the better guide there. Turkmenistan has too few measurements to say anything, and that says something about how hard the country is to measure.

The bridge numbers move with these events. Bridge use hit its four-year high on 21 February 2026, at 272,142, between Iran's two shutdowns. The highest monthly average of 2026 came in June, at 230,635, the month after Iranian connectivity returned around 26 May. Our bridges guide covers which transport to use in each of these countries.

Which bridges and transports people use

Source: Tor Metrics bridge-users-by-transport CSV. Bridges don't report requests per transport, so Tor estimates these by splitting each bridge's requests in proportion to its unique addresses per transport.

obfs4 is the oldest transport in the table, and it still carries almost half of all bridge users. WebTunnel is the one to watch. It carries one bridge user in six, and Tor has been making it the default suggestion in countries where the older transports fail. meek is slow and is now a last resort, used by about 3,400 people a day.

The network: relays, exits, bridges and bandwidth

The relay types in that table do different jobs, and the differences matter when you read the numbers:

Guard

The first relay your connection enters. It's the only part of the network that sees your real IP address, so it's the relay type users depend on most.

Exit

The last relay, which connects to the ordinary website for you. Websites see the exit's address, not yours, so any complaint about Tor traffic goes to whoever runs the exit. Under a third of relays take that role.

Bridge

An entry relay that, unlike ordinary relays, isn't listed publicly, so a censor can't simply block its address.

The three-relay design behind these roles is explained in our guide to what Tor is and how it works.

The network is a little bigger in bandwidth than a year ago and a little smaller in relay count. The October 2026 numbers come with a caveat: exits fell from 3,013 to 2,715 in a single day, 5 October 2026, and total relays slid from about 9,450 to under 8,900 over the following two weeks. A one-day fall that size looks like one large operator's relays going offline together rather than a general decline, and Tor Metrics hasn't logged an event for it. It's worth watching, because the latest-day figures in the table are lower than the 30-day means for that reason.

Relay count also hides how concentrated the network is. In a snapshot of the 9,065 running relays published on 21 October 2026, relays sat in 77 countries and 934 networks. But Germany held 28.2% of the consensus weight (roughly, the share of traffic relays are chosen for), the Netherlands 19.5% and the US 13.9%. For exits, the Netherlands alone carried 31.9% of the weight. The single biggest hosting network, OVH, held 12.7%. Linux runs 88.7% of relays and BSD 10.8%, and Windows accounts for 24 relays in the whole network.

This is where the user count and your own safety meet. Traffic correlation, where one party watches both ends of a connection, gets easier when a few countries and hosting companies carry most of the traffic. Our page on whether Tor can be traced covers what that means for you.

How many onion sites there are

Tor Metrics estimates about 658,000 unique v3 onion addresses a day, as a 30-day mean, with 651,751 on 19 October 2026. A year earlier the figure was around 946,000.

That isn't a count of websites. It's an extrapolation from the onion-service directories (the relays that store onion addresses), made when at least 1% of those relays report their statistics. Every address that publishes itself is counted, whether it's a news site, a chat client's temporary address, a test service or something with nothing on it. The number of onion sites a person could actually browse is a small and unknown fraction of it, and no public list comes close to covering them. The onion sites directory takes the opposite approach: a short list where every address comes from the operator's own source.

Onion services carried about 13.9 Gbit/s of traffic on average over the same 30 days. That's roughly 3% of the 524 Gbit/s the whole network moved. Most Tor use is ordinary websites reached through exits.

If you still see a figure based on the old v2 addresses, ignore it. That series has been noise since 2023, because v2 onions no longer work.

How fast Tor is today

Tor runs its own speed tests, called OnionPerf, from servers in Germany, Hong Kong and the US. These are medians for 23 August to 21 September 2026, the latest the published data reaches:

Where you are matters more than anything you can change. The 1 MiB fetch took a median 1.3 seconds from Germany and 3.6 seconds from Hong Kong. That fits the relay map above, where Germany and the Netherlands hold almost half the network's weight. Onion services take a bit over twice as long as ordinary websites at every size. For reading pages, Tor in 2026 feels slow rather than broken. Video and large downloads are where you notice it.

"Millions of users": how the Tor Project's claim fits the data

Does Tor have 4 million users? The Tor Project itself doesn't put a number on it. The project's own history page says the network has "thousands of relays run by volunteers and millions of users worldwide", and its donation page speaks of "millions of people". Its homepage and press page give no user number at all.

"Millions" fits the Metrics data well: 3.54 million at once on an average day, including bridge users. A round "4 million" is also defensible for many months of 2023, 2024 and early 2026, though not for the most recent 30 days. The honest version is "about 3.5 million at any given time, and more distinct people than that, by an amount nobody can measure".

Downloads give a rough sense of how many different people use Tor. Over the 12 months to 5 March 2026, the last date in the published series, Tor Browser was downloaded at least 21.7 million times from the Tor Project's site, not counting updates. That total misses 14 days that have no data. Downloads aren't people either, since one person can download several times and some downloads are never installed. Windows accounted for 64% of the most recent month's first-time downloads.

Downloading and reusing the data

Everything above came from public sources you can download yourself:

• Tor Metrics: almost every graph has a CSV behind it. Take the graph's address, change .html to .csv, and set start and end dates. The full list of files and their columns is on the statistics page. Skip the comment lines at the top that begin with #.
• Relay details (countries, networks, flags) come from Onionoo, the JSON service behind Tor's Relay Search.
• OONI publishes every measurement, and its API lets you total them by test, country and date, as the censorship table does.
• The method behind every estimate is on the Reproducible Metrics page. Read it before quoting a number.

Our working rules for this page, in case you reuse it: an average means the mean of the days that have data. A "30-day" figure ends on the latest day available. Direct and bridge users are added together only where we say so.