Where Claude Is Available, Country by Country

Where Claude is available, and how to check your country

Anthropic publishes the list of countries and regions where Claude is sold, and that page is the only version of the list worth reading. Check your own country there before you plan anything else. The list lives at anthropic.com/supported-countries, and the Claude help centre carries the same list in article form.

We do not copy the list into this page on purpose. Countries get added and terms get revised. A pasted copy looks authoritative for about a month and then quietly misleads people, because nothing on a static page tells you it has gone stale. Both lists are long, and most of the world is on them, so for most readers the question is answered in one page view.

Read your entry closely rather than scanning for the country name. Some entries are qualified inside the country. Ukraine, for example, appears with the Crimea, Donetsk, Kherson, Luhansk and Zaporizhzhia regions excluded. Anthropic also reserves the right to refuse service to companies whose majority direct or indirect ownership is attributable to a country outside the list, so a business registered in a supported country is not automatically eligible.

Two lists on one page: Claude.ai and the API

The supported-countries page carries two separate lists. One covers the commercial API, the thing you call from code with an API key. The other covers Claude.ai, which is the consumer product: the web app, the mobile apps and the paid Pro and Max plans. The two lists are close to identical, but they are maintained separately because they are two products under two contracts.

Check the list that matches what you want to buy. If you plan to write software against Claude, the API list is the one that decides your answer, and the account you create is a developer console account rather than a Claude.ai subscription. Our guide to how Claude API keys and authentication work covers what that account gives you once you have one.

What "not supported" actually blocks

Sign-up and payment are the gate. An unsupported country means Anthropic will not open an account for you and will not take your money, so the block lands before you ever send a prompt. It is a commercial and legal boundary rather than a technical one. The model does not inspect your location. The account system and the payment flow do.

The terms say this in plain words. Anthropic's consumer terms state that "You may access and use our Services only in compliance with our Terms, including our Acceptable Use Policy, the policy governing the countries and regions Anthropic currently supports". A separate export-control clause asks you to represent that you are "not located in any such country" for embargoed destinations. Those are contract terms, which means the consequence of breaking them is an account action rather than an error message.

You can watch the same boundary operate on a cloud platform, where it is documented with exact error strings. Microsoft's documentation for Claude in Microsoft Foundry lists Offer not available in your country/region and Marketplace purchase eligibility check failed as the two errors you get when your Azure billing account sits in a country the model provider did not open. The compute is there and the model is deployed. The purchase is refused. That is the shape of the whole problem.

Plans and prices, if you are in a supported country

This part only matters once the answer above is yes, so keep it short. As of September 2026, Claude Pro is listed at $20 per month billed monthly, or $17 per month if you pay $200 up front for the year. The Max tiers start at $100 per month. Prices are quoted in US dollars, so your card issuer sets the rate you actually pay.

Our pricing pages cover the rest. Start with which Claude plan matches the work you do if you are choosing a tier. If your interest is the coding tool, read what Claude Code costs per month in practice and then how that price lands once local tax and currency are applied. For the paperwork side, see VAT and invoices on a Claude subscription.

Why we do not publish ways around country rules

We do not publish methods for buying a product in a market where the vendor has chosen not to sell it. That is a house rule and it is not up for discussion here.

The reason is practical as well as contractual. An account whose registered country does not match the terms is an account the vendor can close, and everything inside it goes at the same time: your conversation history and every API key you issued from it. Building a team workflow on that footing means building on something that can be removed without notice. The two options below are duller. They also keep working.

Option one: buy Claude through a cloud provider

Claude is resold on Amazon Bedrock and on Google Cloud Agent Platform, which was previously called Vertex AI. Microsoft Foundry sells it too, in an Azure-hosted version and a version that runs on Anthropic's own infrastructure. This is a genuinely different purchase, because you are buying from Amazon, Google or Microsoft under a billing relationship you already have, and their commercial footprints are not identical to Anthropic's.

Read this carefully, because it is not a loophole. Two gates apply at once. The first is the cloud provider's own rules. Microsoft requires a paid Azure subscription whose billing account sits in a country where the offer is available, and it documents several subscription types that cannot buy Claude at all. Cloud Solution Provider subscriptions are excluded, and so is any subscription without an active pay-as-you-go billing method, which rules out student and free-trial accounts. The second gate is Anthropic's supported regions policy, which Microsoft's own page says still applies on top. Passing one gate does not pass the other.

Where inference runs is a third question again. On Microsoft Foundry as of September 2026, the Azure-hosted Claude deployments are offered across a set of United States regions plus swedencentral in Europe, while the Data Zone Standard deployment type, which keeps processing inside one data zone, is offered for the United States only. Amazon Bedrock splits the same idea into per-region routing and global routing, and publishes a table per model.

Check availability with the vendor's own tooling rather than with a blog post. On AWS:

aws bedrock list-foundation-models --region=us-west-2 --by-provider anthropic --query "modelSummaries[*].modelId"

That prints the Anthropic model IDs your account can see in that region. An empty result means either the models are not offered there or your account has not been granted access yet, which you request from the Model access page in the Bedrock console. AWS keeps the authoritative table on its regional availability by models page.

On Google Cloud, authenticate first, then send one request:

gcloud auth application-default login

MODEL_ID=claude-opus-5
PROJECT_ID=MY_PROJECT_ID

curl https://aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/global/publishers/anthropic/models/${MODEL_ID}:rawPredict \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  -d '{
    "anthropic_version": "vertex-2023-10-16",
    "messages": [{"role": "user", "content": "Hey Claude!"}],
    "max_tokens": 100
  }'

A JSON response containing a content array means the project is enabled and the model is reachable from your account. Note the two differences from the direct API: the model name sits in the URL instead of the request body, and anthropic_version is sent in the body with the fixed value vertex-2023-10-16.

The cloud route is real, and it is how most organisations in this position get access. It is also enterprise-shaped. You need a cloud account with a payment method that survives the provider's own eligibility check, so if your underlying problem is that no international vendor will accept a payment from your country, this route may not solve it either.

Option two: run an open model on a server you rent

This is the answer that does not depend on anybody's sales map. Open-weight models are distributed as files. You rent a Linux server, download the weights and run inference yourself. No vendor has to approve your country, because no vendor is selling you anything.

Ollama is the shortest path to something working. The official Linux installer sets up the binary and a systemd service:

curl -fsSL https://ollama.com/install.sh | sh

Pull a model and talk to it:

ollama pull qwen3:8b
ollama run qwen3:8b

ollama ls lists what you have downloaded, and ollama ps shows what is loaded into memory right now. If ollama run exits immediately instead of giving you a prompt, check systemctl status ollama first, because the command-line client needs the background service to be running.

Size the server by the weights rather than by the CPU count. An 8B model at 4-bit quantisation, which means the weights are stored at four bits each instead of sixteen to shrink the file, still needs several gigabytes resident while it answers, and the whole file has to fit on disk first. Ollama stores models under /usr/share/ollama/.ollama/models on Linux, so that is the filesystem that needs the room. Point it elsewhere with the OLLAMA_MODELS environment variable, and give the ollama user ownership of the new directory with sudo chown -R ollama:ollama /srv/models, because the service runs as that user and cannot write to a directory it does not own.

Quality is the honest caveat. An 8B open model is not a frontier model. For summarising, drafting and code completion it earns its keep. For long multi-step agent work it will disappoint you, and prompt tuning does not close that gap. We compare the trade in choosing an open model instead of a Claude plan, and whether Claude itself can be self-hosted answers the question people ask next. The short version is that Anthropic does not publish the weights, so the local option is always a different model.

Keep the model server private

Ollama binds to 127.0.0.1 port 11434 by default, and that default is protecting you. The API ships with no authentication, so anything that can reach the port can run jobs on your server and read your prompts.

If you only need it from your own laptop, leave the binding alone and forward the port over SSH:

ssh -N -L 11434:localhost:11434 you@your-server

If you genuinely need it on a network, set the bind address through a systemd override rather than editing the shipped unit file, because a package upgrade overwrites the shipped file:

sudo systemctl edit ollama.service

[Service]
Environment="OLLAMA_HOST=0.0.0.0:11434"

sudo systemctl daemon-reload
sudo systemctl restart ollama

Then put a firewall rule or a reverse proxy requiring a token in front of it. Leaving port 11434 open to the public internet with no gate means anyone who scans that port can run work on the server you are paying for. The spending side of that is covered in keeping agent costs predictable on a VPS.

Which option fits you

If you can buy directly, buy directly. It is the least work and the most capable model per unit of spend.

If your country is not on Anthropic's list, but your employer holds a cloud account that passes its provider's eligibility check, the cloud route gives you the same models under a contract that already exists.

If neither applies, rent a server and run open weights. You get a smaller model and more maintenance. You also get something no policy change on a page you do not control can take away from you.

FAQ

How do I check whether Claude is available in my country?

Read Anthropic's supported countries page at anthropic.com/supported-countries. It carries two lists, one for the commercial API and one for Claude.ai, so check the one that matches the product you want to buy. Read your country's own entry rather than scanning for the name, because some entries carry exclusions for specific regions inside the country.

What exactly stops me signing up from an unsupported country?

Account creation and payment. Anthropic's consumer terms limit use of the service to the countries in its supported regions policy, so the restriction is contractual and it is applied when you register and when you pay, not when you send a prompt. The model is not checking where you are. The account and billing system is.

Can I use Claude through AWS, Google Cloud or Azure instead?

Yes, where that cloud provider will sell it to your billing account. Claude is resold on Amazon Bedrock, on Google Cloud Agent Platform and in Microsoft Foundry. Two gates apply: the cloud provider's own country and subscription rules, and Anthropic's supported regions policy on top of them. Microsoft documents the refusal as either Offer not available in your country/region or Marketplace purchase eligibility check failed, and both are raised at purchase time rather than at deployment.

What is the best option if no vendor sells to my country?

Run an open-weight model on a Linux server you rent. Installing Ollama and pulling an 8B model gives you a working local API in a few minutes, and nobody needs to approve your country because nothing is being sold to you. Expect lower quality than a frontier model, particularly on long multi-step tasks, and size the server by the model file rather than by CPU count.