Short answer. The Tor Project estimates that about 3.44 million people are connected to Tor at any moment on an average day. That is the 30-day average to 9 October 2026: 3,213,040 connecting directly to relays plus 225,229 through bridges. A year earlier (10 Sep 2025 to 9 Oct 2025) the estimate was swollen by an unexplained surge. So the 66% fall since then says little about real use.
- Top country: The United States, with 609,239 estimated users a day (19% of relay users), then Germany and the United Kingdom.
- Bridges as a censorship signal: 80% of Tor users in Russia connect through bridges (hidden entry points), against 7% worldwide.
- The "dark web": about 665,086 unique v3 onion addresses a day, an estimate that counts addresses, not websites.
The table below collects the headline figures. Each one is explained, with its caveats, in the sections that follow.
| At a glance | Latest figure | Data period |
|---|---|---|
| Estimated users, direct (relay users) | 3,213,040 | 30-day average, 10 Sep 2026 to 9 Oct 2026 |
| Estimated users, via bridges | 225,229 | 30-day average, 10 Sep 2026 to 9 Oct 2026 |
| Latest single day | 3,597,561 relay + 226,785 bridge users | 9 Oct 2026 |
| Running relays / bridges | 9,347 / 2,362 | 30-day average, 10 Sep 2026 to 9 Oct 2026 |
| Relays with the Exit flag | 3,066 | 30-day average, 10 Sep 2026 to 9 Oct 2026 |
| Advertised / used bandwidth | 1,241 / 543 Gbit/s | 30-day average, 10 Sep 2026 to 9 Oct 2026 |
| Unique v3 onion addresses | 665,086 | 30-day average, 10 Sep 2026 to 9 Oct 2026 |
How many people use Tor in 2026?
Tor Metrics is the Tor Project's own statistics site. It estimates 3,213,040 directly connecting users and 225,229 bridge users per day, averaged over 10 Sep 2026 to 9 Oct 2026 Tor Metrics, directly connecting users (October 2026). Added together that is about 3.44 million. On the latest day with data, 9 Oct 2026, the relay estimate was 3,597,561.
These are estimates, not head counts. Tor deliberately does not collect who its users are. In the Tor Project's words:
"These estimates are derived from the number of directory requests counted on directory authorities and mirrors." Tor Metrics, directly connecting users (October 2026)
"We simply divide directory requests by 10 and consider the result as the number of users." … "The result is an average number of concurrent users, estimated from data collected over a day. We can't say how many distinct users there are." Tor Metrics, Reproducible Metrics (October 2026)
So the figure means "people online at the same time on an average day", not "people who used Tor this year". Those two numbers can differ a lot.
Show the numbers behind this chart (monthly averages)
| Month | Average per day | Days with data |
|---|---|---|
| Oct 2026 | 3,222,555 | 9 |
| Sep 2026 | 3,185,140 | 30 |
| Aug 2026 | 3,118,673 | 31 |
| Jul 2026 | 2,943,920 | 31 |
| Jun 2026 | 3,648,128 | 30 |
| May 2026 | 3,885,004 | 31 |
| Apr 2026 | 3,937,594 | 11 |
| Mar 2026 | 4,287,427 | 26 |
| Feb 2026 | no data | 0 |
| Jan 2026 | 7,152,053 | 31 |
| Dec 2025 | 6,995,972 | 31 |
| Nov 2025 | 7,279,623 | 30 |
| Oct 2025 | 8,128,806 | 31 |
| Sep 2025 | 13,255,961 | 30 |
| Aug 2025 | 1,975,666 | 31 |
| Jul 2025 | 2,143,009 | 31 |
| Jun 2025 | 2,154,056 | 30 |
| May 2025 | 4,088,502 | 31 |
| Apr 2025 | 3,435,587 | 26 |
| Mar 2025 | 2,288,323 | 31 |
| Feb 2025 | 2,167,528 | 28 |
| Jan 2025 | 2,123,665 | 31 |
Blank stretches in the chart are periods with no published estimate: 1 Feb 2026 to 1 Mar 2026; 5 Apr 2026 to 17 Apr 2026. The method skips any day when fewer than 10% of relays report, as "too low for a robust estimate" Tor Metrics, Reproducible Metrics (October 2026). Tor Metrics does not say why these particular periods are missing.
What happened in late 2025? The estimate jumped from about 1.98 million a day in August 2025 to a single-day high of 19.37 million on 2 Sep 2025. The rise appeared in many countries at once. Forum users asked whether it was real Tor Forum (14 Oct 2025), and the Tor Metrics events list gives no cause for it Tor Metrics, events list (October 2026).
The events list does record a related change on 28 January 2026: "Release of tor 0.4.8.22. Includes a change to count consensus downloads only after they have completed." Tor Metrics, events list (October 2026) The linked bug report says directory servers miscount fetches that do not complete Tor issue 41192. Those fetch counts are what the user estimate is built from. Counts fell back over the following months. Today's estimate is still 63% above August 2025, and we can't tell how much of that gap is real.
Bridge users moved differently: 225,229 a day now against 118,926 in August 2025 and 147,521 a year earlier (10 Sep 2025 to 9 Oct 2025).
Which countries use Tor the most?
The United States has the most estimated Tor users: 609,239 a day over 10 Sep 2026 to 9 Oct 2026. That is 19.0% of all directly connecting users. Germany (305,353) and the United Kingdom (173,257) follow Tor Metrics, directly connecting users (October 2026). Relays find the country by looking up the connecting IP address, then discard it Tor Metrics, About (October 2026).
| Country | Relay users a day | Share | Bridge users a day |
|---|---|---|---|
| United States | 609,239 | 19.0% | 25,371 |
| Germany | 305,353 | 9.5% | 5,914 |
| United Kingdom | 173,257 | 5.4% | 4,559 |
| Finland | 153,214 | 4.8% | 1,038 |
| Brazil | 144,511 | 4.5% | 1,315 |
| France | 140,103 | 4.4% | 2,881 |
| India | 124,060 | 3.9% | 2,265 |
| Netherlands | 121,824 | 3.8% | 4,292 |
| Canada | 73,663 | 2.3% | 1,558 |
| Italy | 73,204 | 2.3% | 1,304 |
Read country ranks with care. These are raw counts, not per-person rates, and a country with many servers or VPN exits can look bigger than it is. The ranking also moves. In January 2026, during the surge, the top five were Indonesia, United States, India, Germany, Türkiye. A country can also rank low because Tor is blocked there and users switch to bridges, which the next section covers.
Who uses Tor bridges, and what does that say about censorship?
Bridges are Tor entry relays that are not publicly listed, so they are harder to block. Pluggable transports disguise the connection to a bridge as other traffic. Tor Metrics counts bridge users the same way as relay users, from directory requests that bridges see Tor Metrics, bridge users (October 2026). Over 10 Sep 2026 to 9 Oct 2026 that was 225,229 users a day.
Snowflake (72,222 a day) routes through volunteers' browsers. obfs4 (101,237) makes traffic look random, and WebTunnel (34,553) makes it look like ordinary HTTPS web traffic. Tor Metrics says bridges tell clients apart "by transport protocol, which may include pluggable transports" Tor Metrics, bridge users by transport (October 2026). A per-transport figure is estimated separately, so the rows need not add up to the bridge total.
A high bridge share is a censorship signal. Where plain Tor works, few people bother with bridges. The table lists countries with at least 2,000 Tor users a day, ranked by the share who use bridges.
| Country | Bridge share | Bridge users | Relay users |
|---|---|---|---|
| Russia | 79.6% | 90,212 | 23,157 |
| China | 55.8% | 2,498 | 1,975 |
| Iran | 24.6% | 23,054 | 70,542 |
| Thailand | 16.2% | 3,281 | 16,987 |
| Latvia | 11.9% | 442 | 3,268 |
| Estonia | 9.3% | 400 | 3,886 |
| Pakistan | 7.3% | 991 | 12,676 |
| Hong Kong | 6.6% | 528 | 7,415 |
Russia stands out: 80% of its estimated Tor users connect through bridges. China (56%) and Iran (25%) also rank high. Tor Metrics' events list records, for example, Russian blocking of Snowflake connections in November 2024 and Iran's internet shutdowns from January 2026 Tor Metrics, events list (October 2026). Small countries can rank high from a few hundred users, so check the absolute numbers too.
How big is the Tor network itself?
Over 10 Sep 2026 to 9 Oct 2026 the network averaged 9,347 running relays and 2,362 bridges, all run by volunteers. Tor Metrics only counts relays the directory authorities mark as "Running" Tor Metrics, Reproducible Metrics (October 2026). Exit relays (the last hop, whose IP address websites see) averaged 3,066.
Advertised bandwidth is what relays say they can carry: 1,241 Gbit/s. Used bandwidth, the average of traffic read and written, was 543 Gbit/s Tor Metrics, CSV documentation (October 2026).
| Measure (30-day average) | 10 Sep 2026 to 9 Oct 2026 | 10 Sep 2025 to 9 Oct 2025 | Change |
|---|---|---|---|
| Running relays | 9,347 | 9,060 | +3% |
| Bridges | 2,362 | 2,148 | +10% |
| Relays with the Exit flag | 3,066 | 2,796 | +10% |
| Relays with the Guard flag | 5,159 | 5,803 | −11% |
| Relays with the HSDir flag | 3,366 | 4,969 | −32% |
| Advertised bandwidth (Gbit/s) | 1,241 | 1,114 | +11% |
| Bandwidth used (Gbit/s) | 543 | 400 | +36% |
The guard and HSDir counts swing because those flags (labels the directory authorities assign each hour) depend on uptime and measured speed.
Show the numbers behind this chart (monthly averages)
| Month | Average running relays | Days with data |
|---|---|---|
| Oct 2026 | 9,281 | 9 |
| Sep 2026 | 9,430 | 30 |
| Aug 2026 | 10,103 | 31 |
| Jul 2026 | 9,876 | 31 |
| Jun 2026 | 9,734 | 30 |
| May 2026 | 9,992 | 31 |
| Apr 2026 | 10,183 | 12 |
| Mar 2026 | 9,552 | 30 |
| Feb 2026 | 9,825 | 28 |
| Jan 2026 | 9,738 | 31 |
| Dec 2025 | 9,475 | 31 |
| Nov 2025 | 9,216 | 30 |
| Oct 2025 | 9,074 | 31 |
| Sep 2025 | 9,045 | 30 |
| Aug 2025 | 9,027 | 31 |
| Jul 2025 | 8,767 | 31 |
| Jun 2025 | 8,337 | 24 |
| May 2025 | 8,606 | 31 |
| Apr 2025 | 8,449 | 28 |
| Mar 2025 | 8,970 | 31 |
| Feb 2025 | 8,187 | 28 |
| Jan 2025 | 7,954 | 31 |
How many onion sites are there?
About 665,086 unique v3 onion addresses a day, averaged over 10 Sep 2026 to 9 Oct 2026 Tor Metrics, v3 onion services (October 2026). A year earlier (10 Sep 2025 to 9 Oct 2025) the figure was 931,558. It peaked at a monthly average of 1,150,012 in March 2026. The series stepped down on 2 May 2026, from 948,124 on 1 May to 669,038 the next day. The Tor Metrics events list gives no reason for that drop Tor Metrics, events list (October 2026).
This is the most honest public answer to "how big is the dark web", but it counts addresses, not websites. Tor Metrics measures "the number of .onion addresses that the relay observed in published onion-service descriptors" Tor Metrics, Reproducible Metrics: onion services (October 2026). The total is extrapolated from the relays that report, "if at least 1% of relays reported these statistics" Tor Metrics, v3 onion services (October 2026).
Show the numbers behind this chart (monthly averages)
| Month | Average addresses per day | Days with data |
|---|---|---|
| Oct 2026 | 644,967 | 9 |
| Sep 2026 | 674,965 | 30 |
| Aug 2026 | 696,151 | 31 |
| Jul 2026 | 717,154 | 31 |
| Jun 2026 | 750,309 | 30 |
| May 2026 | 718,668 | 31 |
| Apr 2026 | 997,774 | 15 |
| Mar 2026 | 1,150,012 | 30 |
| Feb 2026 | 1,013,273 | 28 |
| Jan 2026 | 992,063 | 31 |
| Dec 2025 | 989,584 | 31 |
| Nov 2025 | 986,397 | 30 |
| Oct 2025 | 956,757 | 31 |
| Sep 2025 | 910,301 | 30 |
| Aug 2025 | 844,535 | 31 |
| Jul 2025 | 866,617 | 31 |
| Jun 2025 | 876,612 | 30 |
| May 2025 | 861,782 | 28 |
| Apr 2025 | 814,493 | 21 |
| Mar 2025 | 813,285 | 31 |
| Feb 2025 | 836,086 | 28 |
| Jan 2025 | 806,508 | 31 |
Three caveats from the Tor Project itself. The scaling method "is not entirely accurate" Tor Metrics, Reproducible Metrics: onion services (October 2026). Each reporting relay "randomizes its numbers a little bit before publishing them", to protect privacy Tor Blog (26 Feb 2015). And the count says nothing about what, if anything, is hosted at each address. Traffic through onion-service connections (rendezvous circuits) averaged 14.4 Gbit/s Tor Metrics, CSV documentation (October 2026).
What can't Tor Metrics tell you?
- How many different people use Tor. The figure is an average of people online at once. "We can't say how many distinct users there are." Tor Metrics, Reproducible Metrics (October 2026)
- Who they are or what they do. "We only use public, non-sensitive data for metrics," Tor Metrics says. Relays report users by country, "discarding IP addresses" Tor Metrics, About (October 2026).
- Exact numbers. The method says "it is not critical to estimate exact absolute user numbers"; trends matter more Tor Metrics, Reproducible Metrics (October 2026). Days with too few reporting relays are left out, and some periods have no estimate at all (52 missing days in the past 12 months).
- Where people really are. Country comes from GeoIP. A VPN or cloud server puts a user in the wrong country.
- How much is illegal. Nothing in these statistics measures what people use Tor for, and the onion count includes addresses with nothing on them.
- Clean trends across software changes. A counting change, like the January 2026 fix above, can move the numbers without any change in people.
How can journalists and OSINT researchers use Tor Metrics?
Spot censorship and shutdowns. The per-country file has lower and upper columns. If the estimate falls outside them, "a censorship-related event may have occurred" Tor Metrics, CSV documentation (October 2026). A sudden drop in relay users together with a rise in bridge users is the classic sign of a new block.
Check a claim. When someone says "Tor use exploded in country X", download that country's CSV and compare the same dates a year earlier. Then check the events list for counting changes or outages Tor Metrics, events list (October 2026).
Check an IP address. To see whether an IP in your logs was a Tor relay, use our Tor relay lookup. It queries the Tor Project's relay data.
Work safely. Read the Tor and onion OPSEC guide before visiting onion sites, and the dark web OSINT guide for methods and legal limits.
Quote the data correctly. Say "Tor Metrics estimates" and give the date. Say "concurrent users on an average day", not "users per year". The data is released under CC0, so you can reuse it freely Tor Metrics, About (October 2026).
How we built this page
We download the Tor Project's CSV files from metrics.torproject.org every week and recompute every number on this page. The files are userstats-relay-country, userstats-bridge-country, userstats-bridge-transport, networksize, relayflags, bandwidth, hidserv-dir-v3-onions-seen and hidserv-rend-v3-relayed-cells Tor Metrics, CSV documentation (October 2026). Averages are means of the days with data. "30-day average" means 10 Sep 2026 to 9 Oct 2026; "a year earlier" means 10 Sep 2025 to 9 Oct 2025. Charts show 7-day averages and skip weeks with fewer than four days of data.
The totals of relay plus bridge users are our addition, not a Tor Metrics figure. Country names come from the two-letter codes in the files; "??" (unresolved) is left out of rankings.
Download the summary data (CSV)
What changed
- 2026-10-11 First published, with data to 9 Oct 2026.
Sources read: Tor Metrics pages on directly connecting users, bridge users by transport, v3 onion services, Reproducible Metrics, CSV documentation, events list and About; tor issue 41192, "Dirservers miscount networkstatus fetches when they don't complete"; "Some statistics about onions", Tor Blog, 26 Feb 2015; Tor Support, "Does Tor enable criminal activity?"; EFF, "The Legal FAQ for Tor Relay Operators"; Tor Forum thread on the user surge.
Frequently Asked Questions
How many people use Tor?
Tor Metrics estimates about 3.44 million people connected to Tor at any moment on an average day. That is 3,213,040 directly and 225,229 through bridges (30-day average to 9 October 2026). It is an estimate of concurrent users, made from directory requests. The number of different people who use Tor over a month or a year is higher and unknown.
Which country uses Tor the most?
The United States, with about 609,239 estimated directly connecting users a day (19% of the total). Germany and the United Kingdom come next in Tor Metrics data for 10 Sep 2026 to 9 Oct 2026. These are raw counts by IP-based location, not usage per person.
How many onion sites are there?
Tor Metrics estimates about 665,086 unique v3 onion addresses a day (10 Sep 2026 to 9 Oct 2026 average). That counts addresses seen by onion-service directories, not working websites. Tor Metrics does not measure how many of them are working sites.
Is using Tor illegal?
Laws differ by country, so there is no single answer, and this is not legal advice. For the United States, the EFF Legal FAQ hosted by the Tor Project says Tor is not designed or intended to break the law. Some governments block Tor, as the Tor Metrics events list records, but being blocked is not the same as being illegal. Using Tor does not make an illegal act legal; the Tor Project says criminal intent lies with individuals, not the tools they use.
Does Tor Metrics track individuals?
No. Tor Metrics says it only uses public, non-sensitive data. Relays and bridges count directory requests, discard IP addresses and report only country totals, and some statistics have noise added before publishing. The figures cannot show who used Tor or what they did.
Why did Tor user numbers jump in late 2025?
The relay-user estimate rose from about 1.98 million a day in August 2025 to a peak of 19.37 million on 2 Sep 2025. Tor Metrics gives no cause. On 28 January 2026 it recorded tor 0.4.8.22, which counts consensus downloads only after they complete. The linked bug report says incomplete downloads were miscounted. Treat numbers from that period with caution.
Is Tor use growing or shrinking?
Bridge use has grown: 225,229 a day now against 118,926 in August 2025. Direct users are about 63% above August 2025, before the surge. Part of the 2025-2026 movement may come from counting problems, so the long-run trend is uncertain.
Can I reuse Tor Metrics data?
Yes. Tor Metrics states that its data is freely available under a CC0 no copyright declaration, and its graphs are licensed CC BY 3.0 US. Our summary CSV on this page is built from that data and lists the source file for every row.