KosarevaCloud
Run 31 models from eight vendors on Kosareva Cloud from Plexon, billed in rubles with no VPN to set up. The Claude entries ride the Anthropic Messages protocol so cached prompts and tool calls arrive intact; the rest ride OpenAI Chat Completions. Bring your own key.
Related features
One Key, Eight Vendors
Pick Kosareva Cloud in Settings, paste one key, and 31 models from eight vendors are in the picker. Claude Sonnet 5.5 is already selected. The rest of the Claude family sits beside the GPT-6.1 and GPT-6 lines, three Gemini entries, two from xAI, and open-weight entries from Qwen, Moonshot, Z.ai and Mistral.
- Four Claude entries, Sonnet 5.5 selected by default
- Ten from OpenAI across the GPT-6.1, GPT-6 and GPT-5 lines
- Three Gemini entries and two from xAI
- Twelve open-weight entries from Qwen, Moonshot, Z.ai and Mistral, three each
Claude With Its Cache Intact
Most of the list rides the OpenAI Chat Completions wire. The four Claude entries do not. Plexon sends them over the Anthropic Messages protocol, the same path it uses for Anthropic itself, so the prompt arrives as four blocks with three cache breakpoints instead of one flattened string. Kosareva honours the breakpoints: on a live check the first call wrote 5,298 tokens to the cache and the second read the same 5,298 back.
- One provider in the dropdown, two wire protocols behind it
- Cached input at a tenth of the full price on every Claude entry
- Adaptive thinking, streaming and tool calls checked live on Claude Haiku 5.5
- Tool calls on GPT-6 and a Mistral entry, and reasoning streamed apart from the answer on a Z.ai entry, checked the same day
Rubles In, Dollars for Reference
Kosareva prices every model in rubles per million tokens, and that is what you pay. Plexon shows an approximate dollar figure beside each entry, converted at the Bank of Russia rate of 83.76 rubles per dollar on 16 October 2026, so the picker compares with every other provider in the list. The ruble price is fixed. The dollar figure is a snapshot, not the bill.
- Claude Sonnet 5.5: about $2.58 in, $12.89 out and $0.26 cached, per 1M tokens
- Claude Haiku 5.5 and GPT-6 Luna: about $0.13 in and $0.64 out
- Gemini 3.8 Flash: about $0.97 in and $4.84 out
- The cheapest cached input on the list is about $0.013 per 1M tokens
Built for Customers in Russia
Kosareva Cloud LLC operates from Armenia and sells to customers in Russia. Billing is in rubles, the gateway needs no VPN, and the account dashboard is in Russian. Sign in with an email code, VK ID or Yandex ID, then top up from 50 rubles by Russian bank card or SBP, or by invoice from 1,000 rubles for a company. Each key can carry its own spending cap.
- No VPN to reach the gateway
- Top up from 50 RUB by card or SBP; invoices for companies from 1,000 RUB
- Spending caps per key, set in the dashboard
- Available on every Plexon plan
What Is Not Here, and Why
Plexon carries 31 of the models Kosareva sells, and ran both wires live before listing them: chat, streaming and tool calls on the OpenAI side, a cache write followed by a cache read on Claude. Open-weight entries appear under their vendor on this page and by name in the app. A few rough edges are worth knowing before you paste a key.
- Error messages from the gateway arrive in Russian, and so does the dashboard
- Plexon shows no balance bar for this key; the balance lives in the Kosareva dashboard
- Clone from Built-in makes an OpenAI-protocol copy, so a cloned Claude entry loses prompt caching
- Individual open-weight model names stay off this page on purpose; the picker shows every one
Content Is Not Stored
Kosareva does not store request or response content, and checked its gateway logs on 30 September 2026 to confirm it. What it keeps is billing metadata: key id, model, token counts, cost, time, latency and client IP, held in a Moscow data center for as long as the account exists. Web access logs last 14 days. Most models run on the vendor’s own infrastructure outside Russia. Nothing is used for training, and a DPA is signed on request.
- Prompts and answers are gone once the response is sent
- Billing metadata in Moscow while the account exists; access logs for 14 days
- Most models served from vendor infrastructure outside Russia
- Not used for training; DPA on request
Setting it up
Four steps, or one link
Point Plexon at Kosareva by hand in Settings, or let a link do it and confirm once. Either way the key is pasted in Settings and never carried in a URL.
By hand
In Settings
- 1
Open Settings, AI models
The gear at the bottom of the sidebar, or the Ctrl+, shortcut.
- 2
Pick Kosareva Cloud
It sits in the provider dropdown with the others. Nothing to install and no endpoint to type.
- 3
Paste your key
The sk- key from the API keys section of your Kosareva dashboard. It stays on your machine and is used only for requests to Kosareva.
- 4
Choose a model
Claude Sonnet 5.5 is preselected. Switch any time from the model picker in the chat window.
Prompt caching on the Claude entries, tool calling, streaming, reasoning and image input all work from here. The one thing you will not see is a balance bar; the balance lives in the Kosareva dashboard.
Keys
Get a key from Kosareva
Create an account at kosareva.cloud, open the API keyssection of the dashboard and make a key. Sign-in is by email code, VK ID or Yandex ID, and the first top-up starts at 50 rubles.
- Top up by Russian bank card or SBP; companies can pay by invoice from 1,000 RUB.
- Give the key a spending cap in the dashboard if more than one person will use it.
- The dashboard is in Russian, and so are error messages from the gateway.
- Under the key field in Plexon there is an API key helper command, if you would rather keep the key in your keychain than in a config file.
From a link
One link, one confirmation
Plexon registers a plexon:// handler on Windows, macOS, and Linux. A provider link switches one slot and asks you to confirm before it writes anything. The rest of your setup is left alone, and the provider reloads without a restart.
plexon://provider?name=kosareva&model=claude-sonnet-5-5&mode=execution
A key in the URL is refused: links end up in shell and browser history. Paste the key in Settings, or hand it over through an api_key_helpercommand, shown to you in full before anything is armed.
Under the Hood
In the Provider Dropdown
Kosareva Cloud is one of the twelve providers in the Settings dropdown, on every plan. There is no endpoint to type in, and one key covers all 31 models.
Two Wires, One Provider
Most entries use OpenAI Chat Completions. The four Claude entries use the Anthropic Messages protocol, which is what lets cache breakpoints, thinking blocks and Anthropic-shaped tool definitions travel unchanged.
Priced in Rubles
Prices are set in rubles per million tokens and the bill is in rubles. The dollar figures Plexon shows are a one-day conversion at 83.76 RUB per USD, for comparison only.
Long Context, Images In
Up to 1.3M tokens on the largest Z.ai entry, 1.05M on the GPT-6 entries and 1M on every Claude entry. Most of the list reads images.
No VPN
The gateway is reachable from inside Russia as it is. Sign in with an email code, VK ID or Yandex ID; top up by Russian bank card, SBP or invoice.
Content Not Stored
Request and response content is not kept. Billing metadata stays in a Moscow data center while the account exists, and most models are served from outside Russia.
Your Key, Your Bill
Bring your own Kosareva key. Plexon never pools it, and the bill lands on your Kosareva account, in rubles.
Connects to
Nothing here works alone. See the whole map for how the pieces fit together.
Where it goes
This feeds AI Chat & Modes.
Models from eight vendors on one key, billed in rubles, with the Claude entries keeping their cached prompts.