Claude plans when your payment is declined

What each Claude plan costs

Claude plans run from free to 200 US dollars a month, and when your payment is declined the cause is usually the country your card was issued in rather than the card itself. Anthropic sells Claude only in the regions on its published list. If your country is not on that list, no card fixes it, and there is no supported way around it.

Here is the whole ladder in US dollars per month, checked on 2 September 2026. Prices change, so treat claude.com/pricing as the authority and these rows as a snapshot of that day.

[
  {
    "plan": "Free",
    "usd_per_month": 0
  },
  {
    "plan": "Pro",
    "usd_per_month": 20
  },
  {
    "plan": "Max 5x",
    "usd_per_month": 100
  },
  {
    "plan": "Max 20x",
    "usd_per_month": 200
  },
  {
    "plan": "Team standard seat",
    "usd_per_month": 25
  },
  {
    "plan": "Team premium seat",
    "usd_per_month": 125
  }
]

The Team figures are per seat and the others are per account, so read the bars with that in mind. Enterprise is quoted on a different shape again: a seat price plus usage charged at API rates, so the monthly total depends on what people run. How the Enterprise seat and usage price is quoted covers that case on its own.

What each tier gets you

• Free costs 0 and gives you chat and code generation with the smallest usage allowance.
• Pro at 20 dollars a month raises that allowance and opens the specialised tools, including Claude Code.
• Max 5x at 100 dollars and Max 20x at 200 dollars give 5 times or 20 times the Pro usage, with priority access when demand is high.
• Team covers 2 to 150 seats and adds admin controls and one bill. A standard seat is 25 dollars, a premium seat is 125 dollars, and the premium seat includes the Claude Code developer environment.

If you are still choosing, which Claude plan you actually need walks the decision by workload, and what the Pro limits feel like in daily use is the honest version of the middle tier. For a company, Team plans for a small business covers seats and admin.

Does annual billing change the price?

Yes, on the tiers where Anthropic publishes an annual rate. Paying for the year up front lowers the effective monthly price.

[
  {
    "plan": "Pro",
    "monthly_usd": 20,
    "annual_usd_per_month": 17
  },
  {
    "plan": "Team standard seat",
    "monthly_usd": 25,
    "annual_usd_per_month": 20
  },
  {
    "plan": "Team premium seat",
    "monthly_usd": 125,
    "annual_usd_per_month": 100
  }
]

Pro drops to 17 dollars a month when the year is billed up front. There is a payment consequence to that choice: the annual option is one large charge instead of twelve small ones, and a large foreign charge is the harder one to get past a bank that has never seen this merchant before. If you are not sure the card will work, start monthly and move to annual after one clean payment. Annual against monthly billing does the arithmetic in full.

Why your Claude payment is declined

The checkout gives you a short failure message and nothing more. That is not the site being unhelpful. Anthropic's own help page, Why was my card declined?, says they do not receive detailed information from issuing banks about the reason for a decline, so the detail sits with your bank.

Check these four things, in this order:

1. The billing location. Payments are processed only from supported billing locations, so a card whose origin country is outside that set is declined however much money is on it. This is the reason people spend hours on, because the message looks like an ordinary card error.
2. The billing address. It has to match the address your bank holds. A missing accent mark or a misspelled street name is enough to trigger a decline.
3. 3D Secure. 3DS (3D Secure, the extra confirmation step your bank runs during checkout) has to be completed. If the confirmation window opens and gets closed, the payment fails.
4. The card and the funds. Credit and debit cards are accepted. Third party payment services such as PayPal and Venmo are not, so a balance held in one of those is not a payment method here.

When the address and the confirmation step are both right and the card still fails, call your bank before you contact anyone else. They can see the decline code, and you cannot.

Can I pay if my country is not supported?

No, and this is the part worth stating plainly. Anthropic publishes its supported regions at anthropic.com/supported-countries, with one list for the API and one for Claude.ai, and it reserves the right to withhold service beyond those lists. Being outside them is not an error on your side. The product is not sold to you, so the payment cannot succeed.

One detail catches travellers. The check looks at where your payment method comes from and the billing address attached to it, not at where you are sitting. A card issued in an unsupported country is declined while you are in a supported one, and a card issued in a supported country keeps working while you travel. Where Claude is available, country by country tracks the current lists.

There is no supported route around this, and this guide will not give steps for presenting a location or a payment method that is not yours.

Resellers and shared logins: what the risk actually is

Searches for declined payments lead to sellers offering accounts, shared logins and payment services. Here is the part they leave out.

• A bought account is not yours. Whoever set up the email owns it and can take it back. They can also read every conversation in it, since the login is theirs.
• There is no support path. Anthropic supports the account holder. When a resold account stops working, the seller is your only recourse, and that is usually where contact ends.
• A shared login shares one usage allowance and one history. One heavy user exhausts the limit for everybody else on it.
• Terms breaches end accounts. Termination takes the stored work with it, and there is no refund route for money paid to a third party.

When cost rather than country is the real problem, the supported ways to pay less for Claude cover the annual discount and the free allowance, and dropping to a cheaper plan takes a minute inside billing settings. What the free tier gives you matters here too: in a supported country, the free plan needs no card at all.

Run an open model on a server you rent

The path that does not depend on Anthropic's billing rules is an open weight model. The weights are published for download, you run them on hardware you rent, and the only payment relationship is with your hosting provider. Your bank still has to approve that charge, though hosting companies sell in more countries and accept a wider mix of payment methods than a single AI subscription does.

Ollama is the shortest way to a working server. The install script is the one published on ollama.com.

curl -fsSL https://ollama.com/install.sh | sh
ollama --version
systemctl status ollama --no-pager

The script installs a systemd service and starts it, so systemctl status ollama should print active (running). If it prints inactive (dead), start it with sudo systemctl enable --now ollama before going further. Then pull a model and talk to it:

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

Before you judge the speed, check where the model is actually running:

ollama ps

The PROCESSOR column reads 100% GPU when the whole model fits on the card. A value naming the CPU means part of the model spilled into system memory, so every token is generated far more slowly. Check that line first when generation feels slow, because the fix is a smaller model or a bigger card rather than anything in the config.

One more step before you use it from your laptop. Ollama listens on 127.0.0.1:11434 and has no authentication of any kind. Confirm the bind address with ss -tlnp | grep 11434. Do not move it to 0.0.0.0 on a public IP, because anyone who scans that port then has free use of your server. Tunnel to it over SSH instead:

ssh -N -L 11434:127.0.0.1:11434 user@your-server

What hardware tier gets what quality

Quality tracks parameter count, and parameter count sets both the download and the memory you need. These are the default published sizes for one common open family, checked on 2 September 2026.

[
  {
    "label": "qwen3:8b",
    "download_gb": 5.2
  },
  {
    "label": "qwen3:14b",
    "download_gb": 9.3
  },
  {
    "label": "qwen3:32b",
    "download_gb": 20
  }
]

Size by this rule: the weights must fit in GPU memory with headroom left for the context window and the runtime. The 8B file is 5.2 GB, so it sits comfortably on a 12 GB card. The 32B file is 20 GB and wants 24 GB or more. A heavier quantisation shrinks the file and costs some accuracy, which is the knob to reach for when a model nearly fits.

Roughly, a model up to about 8B handles summarising, drafting, classification and small code edits well. The 14B to 32B range holds longer instructions and writes noticeably better code. Above that you are into hardware most people rent by the hour rather than the month.

Be clear about the trade. The best open weights still trail the strongest hosted models on hard reasoning and on very long context, and you take on the operations work: updates, backups, monitoring, and a GPU bill that usually exceeds the Pro plan. The open model wins on availability and control, not on price. Running Qwen 3 on a VPS covers the sizing and the serving setup step by step, what you can realistically self host compares the current open families, and using an open model instead of a Claude plan weighs the swap as a whole.

Who each path is for

• Supported country, working card, occasional use: stay free, then move to Pro at 20 dollars when the limits start to bite.
• Heavy daily use, especially coding: a Max tier, at 100 or 200 dollars a month.
• A small company that needs central billing and admin controls: Team seats.
• An unsupported country, or a card that keeps failing after the four checks above: an open weight model on hardware you rent.

Both paths are ordinary engineering choices. A subscription buys the strongest model with none of the operations work. A rented server running open weights buys a service that does not depend on a payment your bank keeps refusing.

FAQ

Why was my Claude payment declined?

Four causes cover almost all of them. Your payment method's origin country or billing address is outside the supported billing locations. The billing address does not match what your bank holds, down to an accent mark. The 3D Secure confirmation step was not completed. Or the method itself is not accepted, such as a PayPal or Venmo balance. Anthropic states that it does not receive detailed decline reasons from issuing banks, so your bank is the only place the actual decline code exists.

Can I subscribe to Claude if my country is not supported?

No. Anthropic sells Claude only in the regions listed on its supported countries page, and a payment method whose origin country falls outside that list is declined regardless of the balance on it. No supported workaround exists. The available alternative is an open weight model running on a server you rent, which puts the payment relationship with a hosting provider instead.

Is buying a Claude account from a reseller safe?

No. The reseller controls the account email, so they can reclaim the login and can read everything stored in it. Anthropic supports the registered account holder, which means a resold account that stops working leaves you with no support channel and no refund route. Accounts obtained this way also breach the terms, and termination takes the stored conversations with it.

How much does Claude Pro cost in 2026?

Checked on 2 September 2026, Pro is 20 US dollars a month, or 17 dollars a month with the year billed up front. Max is 100 dollars for the 5x tier and 200 dollars for the 20x tier. Team seats are 25 dollars standard and 125 dollars premium, billed monthly. Confirm against claude.com/pricing before you budget, because these move.

What hardware do I need to run an open model instead?

Match the weights file to your GPU memory with headroom for the context window. An 8B model is about 5.2 GB and runs well on a 12 GB card. A 32B model is about 20 GB and wants 24 GB or more. Run ollama ps after loading a model: a PROCESSOR column showing 100% GPU means it fits, and anything naming the CPU means part of it spilled into system memory and generation will be slow.