Odnoklassniki serves a plain HTTP client the whole public group page: 1,434 words, the member count in the meta description, a canonical and breadcrumb JSON-LD, from a German, an American and a Kazakh exit alike. Rendering the same URL costs 2,073,620 bytes instead of 554,224 and returns the same 1,434 words. We measured it on 28 September 2026. The setting for Odnoklassniki proxies is the cheapest one we sell: any exit, plain HTTP, no browser.
Nobody searching for Odnoklassniki proxies is measuring OK
The US first page for this keyword is a row of per-IP vendors. One pitches an "unlimited number of accounts" from a private server, one is a guide for emigrants reaching OK from abroad, one is an SMM-agency setup with a browser extension and a Russian residential IP, and the rest are Russian-language listings. Then Wikipedia and two ok.ru URLs. The suggest box has no data intent at all: odnoklassniki pronunciation, odnoklassniki profile, proxy ok ru, odnoklassniki spam, and, for reasons only Google knows, ok ru hudsucker proxy.
Not one of the ten results fetches an ok.ru page and says what came back. That is the whole gap. OK is the second social network of the Russian-speaking world, available in 13 interface languages from Armenian to Uzbek, owned by VK, and its public groups, topics and profiles are listed in seventeen sitemaps. For brand monitoring, community research, ad verification and public-data collection across Russia, Kazakhstan and the diaspora, what matters is what the site hands a logged-out client. So we measured it.
The full page arrives without JavaScript, from every exit
We audited the official "Vse OK" group, ok.ru/group/50582132228315, with and without JavaScript from two residential exits, then weighed the same URL over plain HTTP from a third. It is the network's own group, 8,565,921 members according to the meta description at the time of the fetch.
| Fetch | Exit | Status | Bytes | Words | What the head carries |
|---|---|---|---|---|---|
| No JavaScript | Germany | 200 | 563,910 | 1,434 | title, description with member count, canonical, H1, BreadcrumbList |
| No JavaScript | United States | 200 | 554,224 | 1,430 | same |
| No JavaScript | Kazakhstan | 200 | 553,528 | n/a (not audited) | same title and description |
| Rendered | Germany | 200 | 2,073,620 | 1,434 | same |
| Rendered | United States | 200 | n/a | 1,430 | same |
The audit diff is empty in both countries: no title change, no description change, no H1 that appears only in the render, no canonical missing without JavaScript, no content that lives only in the script. The four-word difference between Germany and the United States is a live feed, not a regional version. The response even tells you which view you got: OK sets an x-stateid: anonymGroup header on the logged-out group page, so a pipeline can assert on the header instead of guessing from the body.
What the exit does change is the interface language, and only when OK gets to choose it. The short alias ok.ru/ok, fetched from a German exit, came back with German chrome ("Gruppe bei OK.ru, Schliess dich an, lies und kommuniziere") around the Russian group description; the canonical URL from Germany, the United States and Kazakhstan came back in Russian. Parse the canonical URL, not the alias, and the exit stops mattering.
Rendering buys 1.5 megabytes of nothing
On 28 September 2026 the rendered German fetch was 2,073,620 bytes and took 10.6 seconds; the plain fetch of the same page was 563,910 bytes in 3.4 seconds. Same 1,434 words, same member count, same canonical. The extra 1.5 megabytes are the ad stack, the fonts and the messenger widget that OK loads for a browser, none of which a collector needs. This is the same verdict we gave Mastodon, where the rendered profile was 36 times the size of the API answer, and the opposite of TikTok, where nothing arrives without a browser.
Counting a gigabyte as 10^9 bytes, and using the per-GB prices on our pricing page:
| Fetch | Bytes per page | Pages per GB | Datacenter $0.50/GB | Residential Basic $0.80/GB | Residential Premium $2.20/GB |
|---|---|---|---|---|---|
| Plain HTTP (1,430 words) | 554,224 | 1,804 | $0.00028 | $0.00044 | $0.00122 |
| Rendered (1,434 words) | 2,073,620 | 482 | $0.00104 | $0.00166 | $0.00456 |
A thousand public group pages a day over plain HTTP on the datacenter line is $0.28 a day. The same thousand rendered on residential Basic is $1.66, six times more for the same words. Nothing in nine fetches from three countries asked about our address, so the network you pick is a question of price and country coverage, not reputation. The one thing worth paying for on OK is the exit country when you need a Kazakh or Uzbek edition of a regional page, and that is a per-request flag on the rotating pool, not a different product. If you are sizing the pool first, how much proxy data you need runs the arithmetic from the page weight.
robots.txt allows groups, profiles and topics and names 17 sitemaps
ok.ru/robots.txt is 6,025 bytes, and most of it is Clean-param directives, about two hundred of them, listing tracking parameters (st.cmd, ref, mt_campaign, autoplay) that crawlers should strip. The single User-agent: * block disallows the parts of the site that are private or transactional: /messages/join/*, /joincall/*, /gifts/link/*, /discussions/*, /search/profiles/.*, /albums/.*, /games/.*, /music/search/*, plus anything carrying a jsessionid or a cmd= query. There is no AI-crawler block by name, unlike Instagram, which bans them one by one.
User-agent: *
Disallow: /messages/join/*
Disallow: /discussions/*
Disallow: /search/profiles/.*
Disallow: /albums/.*
Disallow: /games/.*
...
Sitemap: https://ok.ru/sitemap-index-group.xml.gz
Sitemap: https://ok.ru/sitemap-index-group-topic.xml.gz
Sitemap: https://ok.ru/sitemap-index-profile-web.xml.gzThen seventeen sitemap indexes: groups, group topics, group catalog, web profiles, user names, VIP profiles and groups, videos, music artists and albums, hobbies, help, congratulations and collection articles. The page we measured is in the group sitemap. That is OK saying, in the one file every crawler reads, which surfaces are meant to be read from outside: public groups and their topics, public profiles, videos. The disallowed paths are the ones behind an account. Read the first set and stay out of the second and you are doing what the file asks.
The API exists, and it is for applications with a key
OK has a REST API at apiok.ru. The group.getInfo method takes a list of group ids in uids and a fields selector, both required, and the documentation states that a session is optional for internal applications and required for external ones; every call is signed with an application key. It is the right tool if you are building an OK application. It is not a public, unauthenticated read like the one Bluesky exposes, and for reading what a logged-out visitor sees it buys nothing over the 554-kilobyte page that already contains the counters.
Where we stop is the same place the robots file stops. The public group page is what OK serves any anonymous visitor and lists in its sitemap; that is what this article is about. Anything that needs an account, running several of them, promoting groups from bought IPs, mass joining, mass messaging, is that account's contract with OK, and the vendor pages ranking for this keyword sell exactly that. We do not help with it. Public reads, pinned country when the market needs it, no browser: that is the whole job.
The setting that works on Odnoklassniki
- Network: the rotating pool on the datacenter line, $0.50/GB, or residential Basic at $0.80/GB when you need a specific country. Nine fetches from Germany, the United States and Kazakhstan produced nine plain 200s with the full page; reputation was never the barrier.
- Fetch mode:
engine: tls, plain HTTP. Without JavaScript the group page is 554,224 bytes and 1,430 words; rendered it is 2,073,620 bytes and 1,434 words. There is nothing on the other side of the browser. - Country: any, for the data. Pin the exit only when you want a regional interface language: the
/okalias returned German chrome from a German exit and Russian from the United States and Kazakhstan, while the canonical URL was Russian from all three. Fetch canonical URLs and the country becomes a free choice. - When the proxy is not enough: there is no Odnoklassniki collector in our catalogue. For the two-view check use the SEO audit API, which fetched this group both ways from one country in 29.6 seconds for $0.0012, and for clean Markdown or JSON from the page use the web scraping API at $0.0002 per page, without rendering.
- Free tier: every account gets $2 of free API usage per month, which at $0.00028 per plain page on the datacenter line is about 7,200 public group pages before you pay anything.
Plain HTTP, any exit, canonical URLs, and leave the browser off. That is the whole configuration for Odnoklassniki.