Platform access by country
Which countries block Discord, YouTube, Instagram, Telegram and more — and hourly checks showing all 38 platforms reachable through residential IPs in 7 countries.
6 hourly rounds and 1,596 checks from seven countries, plus the Russian exit: the numbers behind our live connectivity panels, with the method and the limits, in a form you can quote.
Reach. 1,584 of 1,596 platform checks from the seven non-Russian countries succeeded in their own round. The other 12 had no success that round (Shopee and LinkedIn Ads from the United Kingdom at 23:01 UTC; LinkedIn, Telegram, YouTube, Shopee, and LinkedIn Ads from Singapore at 23:01 UTC; Telegram from Japan at 00:01 UTC; Telegram from Brazil at 00:01 UTC; Telegram from Japan at 04:07 UTC; LinkedIn and LinkedIn Ads from the United Kingdom at 05:08 UTC); the public data does not record why.
Speed. The median time to first response across the seven countries was 1.4 s. Country medians ran from 0.7 s (Singapore) to 2.6 s (Brazil); platform medians from 0.3 s (Google Ads) to 3.3 s (Shopee). The probe server is in Asia, which favours Asian exits.
Exit diversity. No /24 block appeared twice: each round drew a new sticky exit in each country, 42 different /24 blocks in 42 country-rounds, spread over 31 networks (ASNs) — for example T-Mobile and Charter in the United States, BT and Virgin Media in the United Kingdom, KDDI and SoftBank in Japan.
Russia. The Russian exit produced successes in 3 of the 6 rounds; in those rounds 13 to 19 of the 37 platforms counted for Russia answered (Gemini is not counted there, because Google does not offer it in Russia). In the other 3 rounds the probe's run log shows that the Russian exit failed the identity check, so no platform was checked through it. 18 platforms never answered from the Russian exit: Discord, Facebook, Instagram, LinkedIn, Telegram, Threads, X, YouTube, ChatGPT, Claude, Cursor, ElevenLabs, GitHub Copilot, Etsy, Shopee, LinkedIn Ads, Meta Ads, and Reddit Ads. Our check cannot tell a national block from a service that refuses Russian addresses.
IP databases. We looked up three anycast addresses on ipinfo.io, ipwho.is and ip-api.com: the three services agreed on the country and the network (ASN) for all three addresses, on the region for one and on the exact city for none.
Choose the country your account or store lives in, not the fastest one. Each of the seven countries had all 38 platforms answer in at least 4 of the 6 rounds, and the country medians differ by about 1.8 seconds.
Do not count on keeping one rotating residential IP from hour to hour: with the same credentials, our probe got a different /24 block in every round. If an account must keep one IP for weeks, a static ISP IP does not rotate during its lease.
Give scripts room. The median first response was 1.4 s, but 38 of the 1,584 successful checks took longer than 5 s and the slowest took 9.1 s, so a 5-second timeout would have thrown those real answers away.
Russian exits came and went in these rounds (results in 3 of 6; in 5 of all 8 runs in the window, by the probe's run log). Test your own target with 1 GB, the smallest purchase, before you buy more.
| Item | United States | United Kingdom | Germany | Japan | Singapore | Russia | Brazil | Indonesia |
|---|---|---|---|---|---|---|---|---|
| 17 Oct, 21:59 UTC | 38 of 38 | 38 of 38 | 38 of 38 | 38 of 38 | 38 of 38 | Identity check failed | 38 of 38 | 38 of 38 |
| 17 Oct, 23:01 UTC | 38 of 38 | 36 of 38 | 38 of 38 | 38 of 38 | 33 of 38 | Identity check failed | 38 of 38 | 38 of 38 |
| 18 Oct, 00:01 UTC | 38 of 38 | 38 of 38 | 38 of 38 | 37 of 38 | 38 of 38 | 18 of 37 | 37 of 38 | 38 of 38 |
| 18 Oct, 03:05 UTC | 38 of 38 | 38 of 38 | 38 of 38 | 38 of 38 | 38 of 38 | 13 of 37 | 38 of 38 | 38 of 38 |
| 18 Oct, 04:07 UTC | 38 of 38 | 38 of 38 | 38 of 38 | 37 of 38 | 38 of 38 | 19 of 37 | 38 of 38 | 38 of 38 |
| 18 Oct, 05:08 UTC | 38 of 38 | 36 of 38 | 38 of 38 | 38 of 38 | 38 of 38 | Identity check failed | 38 of 38 | 38 of 38 |
Platforms that succeeded with the round's own timestamp, out of 38 (37 from Russia, where Gemini is not counted). "Identity check failed": even after one retry the probe could not confirm the Russian exit's IP and country, so it checked nothing through it that round (run log). The probe ran 8 times in this window; the runs at 01:03 and 02:04 UTC were not saved and are not counted here. Their run-log lines show 37 or 38 of 38 for each of the seven countries and 18 and 19 for Russia.
| Item | United States | United Kingdom | Germany | Japan | Singapore | Russia | Brazil | Indonesia |
|---|---|---|---|---|---|---|---|---|
| Median | 1.8 s | 1.5 s | 1.3 s | 1.1 s | 0.7 s | 2.4 s | 2.6 s | 1.0 s |
| 90th percentile | 2.7 s | 3.0 s | 3.0 s | 3.3 s | 3.0 s | 2.9 s | 4.0 s | 3.4 s |
| Fastest check | 0.2 s | 0.2 s | 0.2 s | 0.2 s | 0.2 s | 0.2 s | 0.3 s | 0.2 s |
| Slowest check | 6.9 s | 6.3 s | 6.3 s | 6.6 s | 6.6 s | 5.2 s | 6.8 s | 9.1 s |
| Checks counted | 228 | 224 | 228 | 226 | 223 | 50 | 227 | 228 |
All successful checks of all rounds in this report. Time runs from the moment the probe opens its CONNECT request to our gateway until the platform's first response headers arrive. Russia counts only the rounds in which it produced results.
| Item | United States | United Kingdom | Germany | Japan | Singapore | Brazil | Indonesia |
|---|---|---|---|---|---|---|---|
| Social and messaging (11) | 1.9 s | 1.7 s | 1.7 s | 1.0 s | 0.7 s | 2.7 s | 1.1 s |
| AI tools (10) | 1.5 s | 1.4 s | 1.3 s | 1.0 s | 0.6 s | 2.2 s | 0.8 s |
| Marketplaces and stores (10) | 1.9 s | 1.6 s | 1.4 s | 1.3 s | 0.9 s | 2.8 s | 1.0 s |
| Ad platforms (7) | 1.9 s | 1.4 s | 1.3 s | 1.1 s | 0.9 s | 2.6 s | 1.2 s |
Median of every successful check of the platforms in each group, all rounds. Russia is left out: it answered for too few platforms to compare groups.
| Item | Median, 7 countries | Fastest country | Slowest country | From the Russian exit |
|---|---|---|---|---|
| Discord | 1.3 s | Singapore 0.5 s | Brazil 2.3 s | no success |
| 1.2 s | Singapore 0.5 s | Brazil 2.0 s | no success | |
| 1.5 s | Singapore 0.8 s | Brazil 2.6 s | no success | |
| 3.2 s | United States 2.7 s | Indonesia 3.5 s | no success | |
| 1.5 s | Singapore 0.7 s | Brazil 2.5 s | answered in 3 of 3 rounds | |
| 1.1 s | Singapore 0.5 s | Brazil 1.7 s | answered in 3 of 3 rounds | |
| Telegram | 3.2 s | United States 2.6 s | Indonesia 3.5 s | no success |
| Threads | 1.4 s | Singapore 0.7 s | Brazil 2.0 s | no success |
| TikTok | 1.3 s | Singapore 0.6 s | Brazil 2.6 s | answered in 3 of 3 rounds |
| X | 1.2 s | Singapore 0.7 s | Brazil 1.7 s | no success |
| YouTube | 2.9 s | United States 2.3 s | Indonesia 3.4 s | no success |
| ChatGPT | 1.2 s | Singapore 0.6 s | Brazil 1.9 s | no success |
| Claude | 1.2 s | Singapore 0.7 s | Brazil 2.4 s | no success |
| Cursor | 1.3 s | Indonesia 0.8 s | Brazil 1.8 s | no success |
| ElevenLabs | 1.3 s | Singapore 0.5 s | Brazil 2.9 s | no success |
| Gemini | 0.3 s | Singapore 0.3 s | Indonesia 0.6 s | not counted (Gemini is not offered in Russia) |
| GitHub Copilot | 1.5 s | Singapore 1.2 s | Brazil 3.1 s | no success |
| Midjourney | 1.3 s | Singapore 0.5 s | Brazil 2.3 s | answered in 2 of 3 rounds |
| Perplexity | 1.3 s | Singapore 0.6 s | Brazil 2.5 s | answered in 3 of 3 rounds |
| Runway | 1.5 s | Indonesia 0.9 s | Japan 2.2 s | answered in 3 of 3 rounds |
| Suno | 1.2 s | Singapore 0.4 s | Brazil 2.8 s | answered in 3 of 3 rounds |
| AliExpress | 1.5 s | Singapore 0.5 s | Brazil 2.6 s | answered in 3 of 3 rounds |
| Amazon | 1.3 s | Singapore 0.7 s | Brazil 2.3 s | answered in 3 of 3 rounds |
| eBay | 1.5 s | Singapore 0.7 s | Brazil 2.6 s | answered in 3 of 3 rounds |
| Etsy | 1.4 s | Singapore 0.8 s | Brazil 2.1 s | no success |
| Lazada | 1.3 s | Indonesia 0.8 s | Brazil 3.0 s | answered in 3 of 3 rounds |
| Mercado Libre | 1.5 s | Singapore 1.0 s | Brazil 2.3 s | answered in 3 of 3 rounds |
| Shopee | 3.3 s | Japan 3.0 s | Brazil 3.9 s | no success |
| Shopify | 1.4 s | Singapore 1.1 s | Brazil 2.9 s | answered in 3 of 3 rounds |
| TikTok Shop | 1.3 s | Singapore 0.9 s | Brazil 2.4 s | answered in 2 of 3 rounds |
| Walmart | 1.2 s | Singapore 0.6 s | Brazil 2.8 s | answered in 2 of 3 rounds |
| Google Ads | 0.3 s | United States 0.2 s | Singapore 0.4 s | answered in 2 of 3 rounds |
| LinkedIn Ads | 3.2 s | United States 2.6 s | Indonesia 3.6 s | no success |
| Meta Ads | 1.2 s | Indonesia 0.8 s | Brazil 2.7 s | no success |
| Microsoft Ads | 1.6 s | Singapore 0.9 s | Brazil 3.2 s | answered in 2 of 3 rounds |
| Pinterest Ads | 1.5 s | Singapore 0.9 s | United States 2.6 s | answered in 2 of 3 rounds |
| Reddit Ads | 1.5 s | Germany 1.2 s | United States 2.2 s | no success |
| TikTok Ads | 1.3 s | Singapore 0.6 s | Brazil 2.8 s | answered in 2 of 3 rounds |
Fastest and slowest country compare each country's median for that platform. "From the Russian exit" counts the 3 rounds in which Russia produced any result.
| Item | Different /24 blocks | Network operators (ASN) | Exit cities |
|---|---|---|---|
| United States | 6 in 6 rounds | Charter Communications (AS11427); Charter Communications (AS20115); Space Exploration Technologies (AS14593); T-Mobile USA (AS21928); Verizon Business (AS701) | Dallas, Delmar, Kansas City, Los Angeles, Waxahachie, Yonkers |
| United Kingdom | 6 in 6 rounds | British Telecommunications (AS2856); Lightspeed Networks (AS210892); TalkTalk Communications (AS13285); Virgin Media (AS5089); Vodafone (AS5378) | Aberdeen, Cheshunt, London, Nottingham, Peterborough, Stoke-on-Trent |
| Germany | 6 in 6 rounds | Deutsche Telekom (AS3320); Vodafone (AS3209) | Berlin, Frankfurt am Main, Helmstedt, Oldenburg, Paderborn, Stuttgart |
| Japan | 6 in 6 rounds | KDDI (AS2516); Matsusaka Cable-TV Station (AS10019); NTT DOCOMO BUSINESS (AS4713); QTnet (AS7679); SoftBank (AS17676) | Hitachi, Isesaki, Matsusaka, Miyazaki, Saitama, Tokyo |
| Singapore | 6 in 6 rounds | MobileOne Mobile/Internet Service Provider Singapore (AS4773); MyRepublic (AS56300); Singtel Fibre Broadband (AS9506); Starhub (AS55430) | Choa Chu Kang New Town, Sengkang New Town, Singapore, Ulu Bedok |
| Russia | 3 in 3 rounds | Ediniy Operator Svyazi (AS60119); Vimpelcom (AS3216); TelCom (AS50676) | Korkino, Moscow, Ulan-Ude |
| Brazil | 6 in 6 rounds | Claro NXT Telecomunicacoes (AS28573); ISP PREMIUM TELECOM (AS52862); ONLINE TELECOMUNICACOES (AS263327); TELEFÔNICA BRASIL (AS27699); V tal (AS7738); power net telecom eireli-me (AS61744) | Bragança Paulista, Guarulhos, Rio de Janeiro, Santa Bárbara d'Oeste, Varjota |
| Indonesia | 6 in 6 rounds | INDOSAT Internet Network Provider (AS4761); SURYA TEKNIKA PRATAMA (AS64027); Telekomunikasi Indonesia (AS7713); Telekomunikasi Selular (AS23693) | Banjarmasin, Batam, Bogor, Medan, Surabaya |
Network names and cities as the probe's IP lookup reports them for each round's exit (ipinfo.io, or ipwho.is when ipinfo.io does not answer); the probe publishes only the first three octets of an IP.
| Item | ipinfo.io | ipwho.is | ip-api.com |
|---|---|---|---|
| 8.8.8.8 (Google Public DNS) | Mountain View, California, US | San Jose, California, US | Ashburn, Virginia, US |
| 1.1.1.1 (Cloudflare DNS) | Brisbane, Queensland, AU | Brisbane, Queensland, AU | South Brisbane, Queensland, AU |
| 9.9.9.9 (Quad9 DNS) | Ashburn, Virginia, US | San Francisco, California, US | Berkeley, California, US |
| Network (ASN), same for all three services | AS15169 · AS13335 · AS19281 | AS15169 · AS13335 · AS19281 | AS15169 · AS13335 · AS19281 |
| Flags the service added | anycast: true on all three | none | hosting: true on all three; proxy: true on 8.8.8.8 |
Free endpoints, looked up at 00:03 UTC on 18 October 2026 from a network in Singapore. These are anycast addresses served from many places at once, the hardest case for any location database.
The probe runs once an hour on our server in Asia (a systemd timer with up to two minutes of random delay). For each of the eight countries it uses rotating residential credentials from our own QA account and the public customer entry proxy.proxy302.com, port 2222, with one sticky session per country.
Before and after the platform checks it asks ipinfo.io through the proxy for the exit IP and country (falling back to ipwho.is). A country's round counts only if the exit is in the right country and the IP did not change; if the exit did not answer the lookup or rotated mid-run, the country is tried once more.
Each platform has one target and one rule for a normal answer, kept in the site's target list: for most sites a 200 on robots.txt, for Telegram a 200 from its web app, for ChatGPT and Claude a 401 from their APIs (a key is missing, the location is accepted; a ChatGPT answer naming an unsupported country does not count), for Discord a gateway address starting with wss://, for GitHub Copilot a 400 or 401 that asks for authorization, and for TikTok Shop and Reddit Ads a 200 or a redirect.
Timing starts when the probe opens the CONNECT tunnel to our gateway and stops at the platform's first response headers, so it includes the tunnel, the TLS handshake and the platform's own response. Most requests ask for the first kilobyte only. Four platforms are checked at a time per country and two countries at a time; TLS sessions are reused per platform host across countries.
Only successful checks are published, as one public file per platform with the latest success per country, its time, latency and exit (first three octets of the IP, city, region, ASN and operator). Successes are kept for six hours. For this report we read all 38 files at each round and kept only results stamped with that round's own time. After every run the probe also writes one summary line to its log: successes per country, or why a country was skipped.
This is not a success rate. Failures are not published, and a refused request, a timeout and a skipped exit all look the same: no result. The counts above are successes, not a measured failure rate of the network.
It is a small sample: 6 of the 8 hourly rounds over about 7 hours, one exit per country per round, from one probe in Asia. Measured from Europe or the Americas, the times could differ.
Reachable is not the same as usable. A 200 on robots.txt or a 401 from an API shows that the platform answered this IP; it does not show that a login, a purchase or an account will work.
The probe has exits only in the eight countries listed, none in mainland China or anywhere else, so this report says nothing about access from other places. The Russian exit failed the identity check in 3 of the 6 rounds, so a missing Russian result proves nothing on its own.
The IP-database test covers three anycast addresses at one moment; it shows that databases can disagree, not how often.
The live files are public and change every hour: https://proxy302.com/live/connectivity/<platform>.json, for the 38 platform names in the table above (discord, youtube, chatgpt, amazon and so on). This report is a fixed snapshot of the rounds listed.
Suggested citation: Proxy302, "Residential Proxy Connectivity Report (October 2026): 38 Platforms, 8 Countries", 18 October 2026, https://proxy302.com/guides/proxy-connectivity-report/ . Please quote figures together with the round times and the limits above.
No. It counts checks that succeeded within their own hourly round, from one probe, over 6 rounds. Failures are not published, so the data cannot say how often requests fail or why.
Our probe server is in Asia, so the path to the Singapore exit is probably the shortest. Times measured from another continent could rank the countries differently.
Both checks fetch robots.txt from Google hosts (ads.google.com and gemini.google.com), and both had medians between 0.2 s and 0.6 s in every one of the seven countries. The check times the first response only, not a full page, and we have not measured why these hosts answer faster.
No. Only successes are published, and a country drops out of a round when its exit fails the identity check. For national blocks see our guide to platform access by country.
Each site reads its own database. In our test all three agreed on the country and network of every address, but for no address did all three name the same city; anycast addresses like 8.8.8.8 are served from many places at once.
The 38 JSON files are public at https://proxy302.com/live/connectivity/<platform>.json and list the latest success per country with its time, latency and exit network. Quote this report with its round times.
Product observation · Confirmed on:
Read all 38 files at each round; rounds generated at 21:59 UTC on 17 October, 23:01 UTC on 17 October, 00:01 UTC on 18 October, 03:05 UTC on 18 October, 04:07 UTC on 18 October, 05:08 UTC on 18 October 2026. A result counts for a round only when its check time equals the round time.
Successes only, kept for six hours; no failures or reasons are published.
Product observation · Confirmed on:
Read from the probe source (deploy/connectivity/probe.mjs, systemd timer) and the target list (liveTargets.json) in the site repository on 18 October 2026.
Local example · Confirmed on:
At 00:03 UTC on 18 October 2026, from a network in Singapore (AS4817 Simba Telecom Pte Ltd), we looked up 8.8.8.8, 1.1.1.1 and 9.9.9.9 on the free endpoints of ipinfo.io, ipwho.is and ip-api.com and kept the raw answers.
Official documentation · Confirmed on:
Official documentation · Confirmed on:
Product observation · Confirmed on:
Read on 18 October 2026 from the system journal of the probe server: one line per run with the successes per country, or why a country was skipped. It lists 8 runs between 21:59 UTC on 17 October and 05:08 UTC on 18 October 2026; the runs at 01:03 and 02:04 UTC are the two not saved for this report, and in the 3 rounds without Russian results the Russian exit is logged as skipped because its IP and country could not be confirmed.
Create an account, top up once, then pick rotating residential, static ISP or datacenter IPs and have one running in your browser within minutes.