cocodot
← Back to guides
Local card declined for overseas AI? cocodot: one card + one key
Claude CodeUpdated 2026-08

Claude Code: Pro vs Max vs Metered API, How the Limits Work, and Paying From Anywhere (2026)

Claude Code is an agent that runs in your terminal. It is not sold separately — usage comes with your Claude subscription — so choosing Pro, Max or metered API is really a decision about how much work you can run.

TL;DR: Claude Code is a coding agent that runs in your terminal and IDE: you give it a goal and it reads the repository, edits code, runs tests and tries again when they fail. It is not sold on its own — usage is included in your Claude subscription, so "which tier" is really "how much work can I run". The logic: occasional use, Pro at $20/month; writing code every day with runs that last tens of minutes, Max at $100 or $200/month, which is an order of magnitude more capacity. The only criterion that matters is how often you hit the ceiling mid-window and sit waiting. There is a third path most people miss: Claude Code can be pointed at metered API billing instead of a subscription, and for lumpy usage — three or four intense days a month — metered is usually much cheaper. Outside a few markets the friction is payment: Anthropic takes overseas cards, and the large recurring charge on the Max tier faces tighter risk scoring than a $20 one.

1. What Claude Code is, and how it differs from an AI editor

An AI editor sits beside you while you write. Claude Code is the other shape: you describe a job and it goes and finishes it. It runs in the terminal, reads a whole repository, edits across files, executes commands to run tests, and when they fail it reads the error and goes round again until they pass. That difference decides where each one fits. Detailed work you want to watch line by line suits an editor; work with clear boundaries and automatic verification — filling in tests, implementing an agreed design, a mechanical refactor — is better handed over. A common stable setup is both: the agent runs the main task, the editor is where you review and refine.

2. How usage is counted, and why you get cut off

This is where new users are most confused. Usage is not counted in questions asked; it is counted in tokens consumed, and it refreshes on a rolling time window rather than resetting at the start of a month. That produces a counter-intuitive result: your total for the day can be modest and you can still be blocked mid-afternoon, because you ran three large tasks inside one window. Knowing this tells you how to pace the work — spread the heavy jobs out, and describe what you want properly before it starts. Re-runs caused by a vague brief are the most wasteful thing you can do with the allowance, by some distance.

3. Pro or Max: a number you can actually compute

Do not decide on instinct. Watch for two weeks and record how many minutes per week you spend waiting because you hit the ceiling. Multiply by four weeks and by what your time is worth, then compare against the difference in price — $80/month from Pro to the $100 tier, $180 to the $200 tier. If you are blocked more than about three hours a week, upgrading almost certainly pays for itself; if you rarely touch the ceiling, it is wasted money. There is a subtler signal too: if you have started avoiding big tasks in order to conserve the allowance, you should already have upgraded. At that point the cost is no longer waiting time, it is the work you are not doing — larger, and much easier to miss.

4. The third path: metered API instead of a subscription

As well as signing in with a subscription account, Claude Code can be configured to bill against metered API usage. The cost curves are completely different. A subscription is a fixed cost you pay whether you use it or not; metered is variable and a quiet month costs almost nothing. The deciding factor is how evenly your usage is spread. Heavy every-day use generally favours the subscription, which effectively buys out your peak. But a project rhythm — flat out for three or four days near a deadline, barely touched otherwise — usually favours metered by a wide margin, and it cannot throttle you on the days that matter most. A quick self-test: how many days last month did you genuinely use it hard? More than twenty, subscribe. Fewer than ten, look seriously at the API.

5. The payment barrier

Anthropic bills in USD and accepts overseas cards. Cards issued in many markets are declined at checkout, and the cause is not the card network — platform risk models read the issuing institution's country from the first digits of the card number. That is why trying a different local bank rarely helps: you changed the bank, not the thing being scored. Max adds a second consideration: a recurring $100 or $200 charge is scored more tightly than a $20 one, so the card needs to be reliable rather than merely acceptable, and it needs to be reliably funded. Plan for the charge to be a little larger than the sticker price on any given month, because FX movement and authorisation holds both push in the same direction.

6. Paying with a virtual card, end to end

Without handing your account to anyone: ① sign up at cocodot and top up the wallet, Alipay included; ② issue a US-BIN Visa virtual card and move balance onto it — wallet balance is not card balance — and for the $200 Max tier keep something above $210 on the card so FX movement and temporary holds have room; ③ at claude.ai go to Settings → Billing, choose Pro or Max, and enter the card details with the billing address exactly as your console shows it; ④ sign in to the same account from the terminal and start working; ⑤ note the renewal date and check the card balance two or three days ahead each month, because a lapsed subscription means going through the whole flow again. Current fees are published at cocodot.co/pricing.

7. Troubleshooting

Payment declined — sort by symptom first: declined immediately is usually the issuing country; works elsewhere but fails here points at a billing address that does not match the issuer's record; first month fine and second month declined is a short balance on the renewal date. Even a good card now fails — the account has been flagged by repeated attempts; stop for a day or two, because continuing makes it worse. Subscribed but the terminal will not work — check whether you signed in with the subscription account or configured an API key; these are two independent billing systems, and mixing them produces the confusing "I am subscribed and it says I have no credit". It stops mid-run — you have hit the window's usage ceiling; wait for the refresh, or break the task up and brief it more precisely to cut down re-runs.

Three ways to pay for Claude Code

Pro $20/moMax $100 / $200Metered API
SuitsLight use, occasional small tasksDaily coding, long continuous runsLumpy usage, or wiring it into your own pipeline
CapacityYou will hit the ceiling quicklyAn order of magnitude moreNo ceiling — you pay for what you use
Hitting the limitCommon; wait for the window to refreshUncommonNo allowance window; request-rate and concurrency limits still apply
Cost in a quiet month$20 regardless$100-200 regardlessClose to zero
Callable from your own codeNoNoYes
Payment difficulty abroadOverseas card; $20 clears more easilyOverseas card; large charges scored harderBilling account, or a relay taking local payment

FAQ

Do I have to buy Claude Code separately?

No. Usage is included in a Claude subscription — Pro or Max both give you access, and it is not sold on its own. Choosing a tier is therefore really choosing how much work you can run.

Is Pro enough, or do I need Max?

It depends entirely on how often you hit the ceiling. Occasional small tasks are fine on Pro; daily coding with long runs will block frequently. Put a number on it: multiply your weekly waiting time by what your time is worth and compare with the price difference. More than about three hours a week and upgrading usually pays.

Why am I rate limited when my usage is not high?

Because usage is measured over a rolling window, not per day or per month. Running several large tasks inside one window hits the ceiling even if your daily total is modest. Spreading heavy work out and briefing tasks clearly to avoid re-runs both help noticeably.

Subscription or metered API — which is cheaper?

It comes down to how even your usage is. Steady heavy use favours the subscription. A project rhythm, where a month contains three or four intense days, usually favours metered billing by a wide margin, and it will not throttle you during those days.

Why do the larger Max charges fail more often?

The bigger the amount, the lower the risk threshold on both the issuer's and the merchant's side, and recurring large charges are watched particularly closely. What helps: a US-BIN card, a billing address identical to the issuer's record, and a balance comfortably above the monthly fee rather than exactly equal to it.

I subscribed but the terminal says I have no credit.

Check what the terminal is signed in as. Subscription billing and API billing are separate systems, so a client configured for API mode without API credit will report exactly this, regardless of an active subscription.

About cocodot

cocodot is a payment and AI access service for developers and cross-border teams in mainland China. It provides US-BIN virtual cards issued by a licensed institution — used to pay for overseas subscriptions and ad accounts — and an OpenAI-compatible AI API gateway for calling Claude, GPT and Gemini from within mainland China. Both share one wallet, funded by Alipay and accounted in USD. Card: $9.9 to open, 3% to load, $1 per active card per month; spending: $0.60 settlement fee on purchases under $20; a corresponding fee applies when the issuer charges one.

Service scope, pricing and limits →
Claude Code: Pro vs Max vs API, and How to Pay · cocodot