Claude Code Subscription vs API Billing: Costs, Setup, and Choice

Claude Code is installed, but the payment decision is still unclear: should you sign in with a Claude subscription, fund the official API, or connect an Anthropic-compatible gateway? Those choices can open the same terminal interface while charging different accounts.

Choose the billing route before you paste any credential. This guide shows what each option buys, how current subscription and token prices compare, which regional checks matter, and how to run a small verified BetterToken test without exposing an API key.

Choose the billing route by what you actually need

The quickest answer is scenario-based: use a subscription when you want Claude apps and regular interactive coding; use an API account when you need your own keys, automation, or usage-based billing; avoid buying someone else’s login.

Route	What you pay for	Best fit	Main check before paying
Your own Pro or Max subscription	Access to Claude apps and Claude Code under the plan’s shared usage limits	Individuals who use Claude chat and Claude Code frequently	Your country is supported, the plan fits your workload, and you understand optional usage credits
Official Claude Console API	Model usage billed separately at API rates	Scripts, CI, products, team integrations, or intensive bursts that need API keys	Console eligibility, model pricing, spend limits, and auto-reload settings
Third-party Anthropic-compatible gateway such as BetterToken	Usage on the provider’s own balance, prices, key, and endpoint	Developers who need an alternative billing relationship or a compatible endpoint	Current model availability, price group, payment terms, data handling, and organizational policy
Resold or shared Claude account	Access controlled by a seller rather than by you	No dependable professional use case	Recovery rights, confidentiality, other users’ activity, and interruption risk

A paid Claude plan and the Claude Console are separate products and separate bills. BetterToken is another separate account: topping it up does not activate Claude Pro, and buying Pro does not create BetterToken credit.

How much does a Claude subscription cost?

The published web subscription prices checked on September 30, 2026 are:

Plan	Published price	What to remember
Pro	US$20 billed monthly, or US$200 paid upfront annually	The annual option works out to about US$16.67/month; the pricing page rounds this to US$17. It is not a US$17 monthly charge.
Max 5x	US$100/month	Higher usage allowance, not unlimited use.
Max 20x	US$200/month	Higher allowance again, with applicable usage limits.

Applicable taxes are additional, and mobile-store prices can differ. Consult current pricing and Max billing details before purchase. Team and Enterprise users should check their organization’s seat and billing arrangements rather than assume an individual-plan price applies.

You do not need to buy a separate personal “Claude Code subscription” on top of an eligible Pro or Max subscription. However, the subscription allowance is shared across Claude and Claude Code, and it does not include a Console API balance. Optional usage credits or a switch to API billing can introduce additional spending; review those settings before accepting them. See Claude Code with Pro or Max.

Which login option pays the bill?

The login menu is a choice of account and billing route, not simply a choice of interface.

Menu label	What it means
Claude account with subscription	Sign in with your eligible Claude account; Pro, Max, Team, or Enterprise usage follows the applicable plan or organization settings.
Anthropic Console account	Use the official Console account and its API billing. Your personal subscription payment does not fund this balance.
3rd-party platform	Configure a supported cloud integration such as Amazon Bedrock, Microsoft Foundry, or Vertex AI, with the corresponding cloud account.

The authentication documentation explains these routes. The third menu entry is not a universal login for every API gateway. BetterToken uses its documented Base URL and token configuration instead.

For billing, open the service you actually used: Claude.ai billing for a web subscription, Console billing for the official API, or BetterToken Workspace for BetterToken usage. The search term claude.console.api is not a Base URL you should paste into Claude Code.

One common surprise is an old environment variable. Claude Code can select a configured API credential instead of your saved subscription login. In particular, ANTHROPIC_API_KEY is used in non-interactive -p mode when present unless a higher-priority authentication method applies. Check the credential precedence rules, not just whether you previously logged in to Pro.

Check region and payment availability before choosing a card

Service eligibility comes before payment method. Anthropic publishes separate supported-country lists for Claude.ai and its commercial API. Check the current list for the service you intend to use; a foreign card, phone number, or VPN does not by itself change account eligibility.

Taxes, accepted cards, app-store pricing, currency conversion, and verification can vary by account and country. For an official subscription, inspect the final Claude checkout. For the official API, inspect Console billing and auto-reload. For BetterToken, use the methods, currency, exchange rate, fees, minimum, and credited amount shown in the current top-up flow rather than relying on an old screenshot.

A scoped note for users in Russia

As checked on September 30, 2026, Russia was not listed on Anthropic’s supported-country page for Claude.ai or the commercial API. That is a Russia-specific constraint, not a statement about every country. BetterToken’s current Console exposes USD and RUB top-up choices; the exact payment methods and settlement terms are determined at checkout. Start with a small supported amount and verify the balance before testing a request.

Do not solve a billing problem by buying a shared login

A shared or resold account can look cheaper, but you do not control recovery, concurrent use, security settings, or continued access. For source code, use an account and credential you control, then review the provider’s data terms and your organization’s policy.

Connect Claude Code to BetterToken and verify the route

This setup is for a personally managed terminal. It does not override employer policy, and it does not convert BetterToken credit into a Claude subscription.

1. Create your own account, balance, and API key

Open BetterToken Workspace, inspect the current top-up terms, and add only a small test balance. Create your own BetterToken API key and set an expiry or optional spending limit where appropriate. The quick-start guide documents the account, balance, and key flow.

Do not use an Anthropic key in this configuration. Do not paste the BetterToken key into Git, a screenshot, a ticket, or a shell-history command.

2. Set the documented endpoint in Bash

The current English and Russian BetterToken guides now use Anthropic Messages at https://bettertoken.ai without /v1; Claude Code appends /v1/messages. Both manual-setup pages also tell you to copy the exact Model ID from the live catalog and set the main and default model variables. For a controlled Sonnet 5 test, run these lines in Bash, paste the key at the hidden read, and press Enter:

export ANTHROPIC_BASE_URL="https://bettertoken.ai"
export CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC="1"
export API_TIMEOUT_MS="3000000"
export ANTHROPIC_MODEL="claude-sonnet-5"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="$ANTHROPIC_MODEL"
export ANTHROPIC_DEFAULT_SONNET_MODEL="$ANTHROPIC_MODEL"
export ANTHROPIC_DEFAULT_OPUS_MODEL="$ANTHROPIC_MODEL"
read -rs ANTHROPIC_AUTH_TOKEN
export ANTHROPIC_AUTH_TOKEN

As verified on September 30, 2026, the exact public ID used here is claude-sonnet-5. The example maps the Haiku, Sonnet, and Opus aliases to that same model so one test cannot silently switch models. For production, copy current IDs from the catalog and map roles separately when needed. If a cached or localized document shows different fields, reopen the live setup page, clear conflicting environment variables, fully restart Claude Code, and confirm the model and charge in Workspace.

These exports apply only to the current shell. For a persistent configuration, merge the same fields into your existing Claude Code settings instead of overwriting unrelated settings, then fully restart the client.

3. Run one constrained request and find its usage record

Start Claude Code in a small test directory:

claude

Send a harmless message such as:

Do not read files or use tools. Reply only: connection ok.

A normal reply without an authentication, connection, or model error is the first success condition. The second is a matching request in Workspace with the expected time, key, model, token usage, and charge. Both checks matter: creating a key or opening the terminal alone does not prove which provider handled the request.

Understand what claude -p changes—and what it does not

-p is the short form of --print: Claude Code runs non-interactively and prints the result. It is useful for scripts, but it is not a free mode and does not choose a billing account by itself.

The active credential decides the bill. ANTHROPIC_API_KEY overrides a saved Pro, Max, Team, or Enterprise login and is used in -p mode only when no higher-priority provider or authentication method applies. With no higher-priority provider active, a configured ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN route the request to the corresponding gateway. Before a cost test, inspect inherited variables and keep only the provider you intend to use.

A minimal test is:

claude -p --max-turns 1 "Do not read files or use tools. Reply only: connection ok."

For an API-billed task, add client-side controls:

claude -p --max-turns 3 --max-budget-usd 1.00 "Read only README.md and explain how to start this project. Do not modify files."

--max-turns limits agentic turns. The current CLI reference defines --max-budget-usd as the maximum API spend before stopping in print mode; the full subagent cap behavior requires Claude Code v2.1.217 or later. Confirm your local version with claude --version, and reconcile the run against the provider’s usage ledger; see the current CLI reference.

Compare subscription and API cost with the same real tasks

A subscription price and a per-token price are not directly interchangeable. A plan also covers Claude apps and has shared usage limits; an API route gives you keys, automation, and metered usage. Compare the cost of completing the same representative work.

As of September 30, 2026, Anthropic’s Sonnet 5 model page identifies the exact API ID as claude-sonnet-5 and marks it active as a legacy model. The public base-token listings showed:

Billing route	Input base tokens	Output base tokens	Scope
Official Claude API list price	US$2.00 per 1M tokens	US$10.00 per 1M tokens	Standard global first-party base price for this exact model
BetterToken public Claude catalog	US$1.36 per 1M tokens	US$6.80 per 1M tokens	Published catalog price for claude-sonnet-5 in the Claude group

The BetterToken figures are 68% of Anthropic’s base rates—32% lower for the same input/output token mix. The current official pages do not support a Sonnet 5 change to US$3/US$15 on September 1; the US$3/US$15 rows in Anthropic’s price list belong to Sonnet 4.6 and 4.5. The table includes the required base input and output charges, not every possible invoice item. Cache writes and reads, server tools, data-residency modifiers, negotiated discounts, taxes, top-up or payment fees, currency conversion, the key’s access group, and actual model routing can change the final amount. Treat the catalog as an estimate, then use the matching request record as the billing check; reopen official API pricing and the BetterToken catalog before paying.

For one run, start with:

estimated cost = input_tokens / 1,000,000 × input_rate
               + output_tokens / 1,000,000 × output_rate
               + cache, tool, and other applicable charges

Then run 5–10 representative tasks and use the provider’s actual records. If you regularly use Claude chat as well as Claude Code, compare the measured API total with Pro or Max—not with the subscription fee alone. If you mostly run CI jobs or occasional isolated tasks, usage billing may be easier to control with per-key limits.

Troubleshooting: follow the payment or request to its destination

Symptom	Check first
You paid for Pro but see an API charge	The active provider, API environment variables, and the account that recorded the request.
You cannot find an API bill on Claude.ai	Open official Console billing or BetterToken Workspace, depending on the request route.
401	The key is valid, copied without extra whitespace, and used with the correct provider.
403 or model unavailable	Account/model permissions, provider availability, and regional or organization restrictions; read the error body.
404 or an endpoint error	Use the documented Base URL and avoid a duplicated /v1 path.
429, timeout, or 5xx	Read the response for rate, quota, or balance details; inspect request history before retrying a potentially completed task.
The login menu still appears	Check that the same terminal inherited the token and that saved settings or cloud-provider flags are not selecting another route.

Consult the BetterToken setup troubleshooting and Claude Code authentication guide. For support, send the error code, request time, and request ID when available—not your API key.

Common questions before you commit

Do I need both a Claude subscription and API credit?

Only when you need both products. A subscription is enough for eligible interactive Claude and Claude Code use within its limits. Console or gateway credit is separate and is useful for API keys, automation, or additional metered usage.

Can BetterToken balance upgrade Claude.ai to Pro or Max?

No. BetterToken, Claude subscriptions, and the official Console are separate billing relationships. Credit cannot be transferred between them.

Is “3rd-party platform” in the Claude Code login menu a universal gateway login?

No. That menu covers documented platform integrations. An Anthropic-compatible gateway is configured with its own Base URL and authentication variables according to that provider’s instructions.

Is an API gateway the same as a shared Claude account?

No. With your own gateway account and key, you control that key rather than borrowing another person’s login. The gateway is still in the data path, so review its terms, retention, support, and your company’s rules before sending private code.

What should I do when my country is not supported by Anthropic?

Do not treat a VPN, foreign card, or rented phone number as a change in eligibility. Recheck Anthropic’s current country list and the terms of any separate provider you consider. A third-party endpoint has its own account and policies; it does not make an unsupported official account supported.

How do I know the test used the intended provider?

Require both a normal terminal response and a matching usage record in the intended billing account. If the record appears elsewhere, inspect environment variables, saved settings, and the selected login before running a larger task.

Make the choice with one small, measurable test

Choose Pro when you are eligible, use Claude apps regularly, and want Claude Code inside one personal subscription. Compare Max only after you know Pro’s shared limits are the bottleneck. Choose the official Console API when first-party API keys, automation, or organizational controls matter. Consider BetterToken when its separate billing route, current prices, and Anthropic-compatible endpoint fit your needs—and only after checking terms, data handling, and regional availability.

Whichever route you choose, begin with one low-risk task, confirm the response and the matching charge, then scale. For the gateway path, review the current Claude Code setup and live model catalog before creating a production key.