MMOOZZGG.com

Neural network API

Claude API for business: status in Russia, data and alternatives

9 min read9 sources

On 16 October 2026, the Claude API is officially not available from Russia: Anthropic does not include Russia and Belarus in its list of supported countries, and its rules also exclude companies controlled from such countries. Below: the models and how pricing works, from Anthropic’s official pages, how long requests are kept, and two lawful paths for business: an official company account where Anthropic operates, or an open model with a compatible API. We give no instructions for getting around the rules.

Access from Russia:Officially not available from Russia(checked 2026-10-16, source)

In short: the status on the date of the check

We checked Anthropic’s official pages on 16 October 2026. The Supported Regions Policy has applied since 12 May 2026 and treats any country outside its list as unsupported. Russia and Belarus are not on the list.

The key points on this date:

  • Models: Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5, all with a context window of 1 million tokens.
  • Prices: in US dollars per million tokens, with different rates for input and output, and discounts for caching and batch processing.
  • Data: API inputs and outputs are deleted within 30 days and are not used for training by default.
  • Access from Russia: closed both for people in Russia and for companies controlled from Russia.
  • What to do: an official company account in a listed country, if the company meets Anthropic’s terms, or an open model chosen by a test on your requests.

Claude models and prices

On 16 October 2026, Anthropic names four current models. Claude Fable 5.1 is built for complex reasoning and long agent work, Claude Opus 5.5 for long programming and knowledge tasks, Claude Sonnet 5.5 combines speed and quality, and Claude Haiku 5.5 serves large flows of simple tasks: classification, data extraction and routing.

All four have a context window of 1 million tokens and up to 128 thousand tokens in an answer. Earlier versions, among them Claude Opus 5, Claude Sonnet 5 and Claude Haiku 4.5, stay available as legacy models. The same models are sold through Amazon Bedrock, Google Cloud and Microsoft Foundry, and Bedrock and Google Cloud have their own prices.

The price is in US dollars per million tokens, with different rates for input and output for each model. Prompt caching costs extra on writes: 1.25 times the normal input price for a 5-minute cache and 2 times for a 1-hour cache. Reading from the cache costs 10% of the input price, and less for the new models: 5% for Opus 5.5 and Sonnet 5.5, 2.5% for Fable 5.1.

Batch processing (Batch API) gives a 50% discount on input and output, and the cache and batch discounts add up. Processing only in the US costs 1.1 times more. For Haiku 5.5, requests longer than 100 thousand tokens are billed at a higher rate; for the other current models, the full 1 million tokens is billed at the normal price.

Watch the tokenizer: models from Claude 4.7 onward count about 30% more tokens for the same text than earlier ones. The bill for the same work can grow even at the same price per token.

Payment on 16 October 2026: by actual monthly use, in US dollars only. For standard accounts, Anthropic accepts major bank cards; credits are bought in the Billing section of Claude Console, and enterprise clients can pay by invoice. Request limits depend on the account tier: Start, Build and Scale; above Scale, terms are agreed with the sales team.

We keep the per-token rates in figures in the API price comparison, and the cloud API cost calculator estimates spending for your volume.

Claude Code and agents through the API

Claude Code works as a programming agent in the terminal. The Agent SDK gives the same tools, the agent loop and context management as a library for Python and TypeScript. The agent plans its steps and calls tools on its own: it reads and edits files and runs commands.

Agents on the Agent SDK work with an API key. Anthropic does not allow third-party developers to build claude.ai sign-in and subscription limits into their products without a separate agreement. Use of the SDK falls under Anthropic’s Commercial Terms of Service.

There is an option where Anthropic itself runs the agent: Claude Managed Agents. The bill has two parts: tokens at the normal rates, and session time at an hourly rate, counted to the millisecond and only while the session runs. The batch discount does not apply to such sessions.

For a Russian company, all of this comes back to the same status: Anthropic’s region rules apply to Claude Code directly. An agent on an open model can be built through a compatible API: a vLLM server exposes interfaces in the OpenAI format, and agent libraries that understand it connect without a rewrite.

How Anthropic stores requests

The Anthropic Privacy Center page, updated on 1 July 2026: API inputs and outputs are deleted automatically on Anthropic’s side within 30 days after they are received or created.

The same page lists the exceptions: services where you control storage, such as the Files API; a separate agreement, such as zero retention; legal requirements; and usage policy violations. If an automatic check flags a request as a violation, inputs and outputs are kept for up to 2 years, and safety classifier scores for up to 7 years.

By default, data from commercial products, including the Anthropic API, is not used to train models (article updated on 18 August 2026). The exceptions: you sent feedback, for example with the rating button, or you allowed the use of the data yourself. Feedback is kept for up to 5 years.

Anthropic gives Zero Data Retention to some Claude Platform and Claude Code for Enterprise clients after approval, deciding for each organization. Even with zero retention, Anthropic keeps the results of its safety classifiers.

For a Russian company, 152-FZ, Russia’s personal data law, also matters. Since 1 March 2023, an operator must notify Roskomnadzor, the Russian data protection regulator, before a cross-border transfer of personal data (Part 3, Article 12), and the data of Russian citizens is collected and stored in databases in Russia (Part 5, Article 18).

Access from Russia

On 16 October 2026, Anthropic’s region policy names three cases of unsupported use: a person is physically in an unsupported region; a company is registered or headquartered in such a region; a company is directly or indirectly mostly owned or controlled by persons from such a region, wherever its staff are.

The rule covers the Claude apps, Claude Code, the API and the developer platform, as well as Claude models on cloud platforms and at authorized resellers. So a key from a cloud or from a middleman does not change the status, and the middleman also sees all your requests.

The official account path does not suit everyone. The company must be registered in a listed country and must not be controlled from an unsupported region, and the people using the service must be physically in a supported country. If the terms are met, we help connect an official account. If not, an open model remains.

The status of other providers on the same date is on the page about AI model APIs in Russia. How the lawful ways to get model access differ is covered on the page about where to buy AI model APIs.

A replacement for large volumes

For large volumes in Russia, a working replacement for the Claude API is an open model, such as DeepSeek or Qwen. You can run it on your own servers or in your own data center, and then requests do not leave the company’s infrastructure.

Anthropic has its own request format, while vLLM serves an open model in the OpenAI format, with the Chat, Completions and Responses interfaces. When you move from Claude, the model call layer changes, while the prompts, tools and agent logic stay.

We choose the model by a test on your requests: we run a sample of real tasks through several open models and compare them with the Claude answers you treat as the reference. This shows which tasks an open model handles at once and where tuning is needed.

For large volumes, we offer such an API as a service: DeepSeek, Qwen and other open-weights models, the OpenAI format, access to the infrastructure and to request processing logs, and your requests are not used to train models. Details in the block below.

Questions and answers

How much does the Claude API cost? On 16 October 2026, Anthropic charges in US dollars per million tokens, with different rates for input and output. Batch processing gives a 50% discount, and reading from the cache costs 2.5% to 10% of the input price, depending on the model.

The rates by model in figures are in our API price comparison.

Can you buy the Claude API in Russia? Not officially: on 16 October 2026, Russia is not on Anthropic’s list of supported countries, and the rules also exclude companies controlled from unsupported regions. We do not resell access to Claude.

We help connect an official company account where Anthropic operates, if the company meets the terms, or we choose an open model by a test on your requests.

Does Anthropic store my requests? Yes, by default for up to 30 days: API inputs and outputs are deleted within this term. If a request is flagged as a policy violation, the data is kept for up to 2 years.

By default, API data is not used for training, and zero data retention is available to some enterprise clients after Anthropic’s approval.

Sources and dates

Contact us