Claude Code pricing 2026: Pro and Max subscription or API key, cached session cost, and how to pay from Indonesia

People usually ask, “Is Claude Code free or paid?” The answer is: the application is free, but the model is paid, and the model can be paid for in two ways that do not overlap.

• Claude subscription. Claude Code is included with Claude Pro ($20 per month, or $17 per month when paid $200 upfront for a year) and Claude Max ($100 or $200 per month). These are Anthropic's US prices before tax. The Free plan is not included.
• API key, pay per token. There is no monthly fee. In Kunavo, Claude Sonnet 5 is priced at $1.40 per 1 million input tokens and $7.00 per 1 million output tokens; a typical agent step (roughly 25,000 input tokens and 1,200 output tokens) costs approximately $0.0434.
• How you pay. Web subscriptions accept only credit or debit cards. Kunavo balance can be topped up with Visa or Mastercard cards (credit or debit), Link, or Google Pay, starting at $10. QRIS and e-wallets such as GoPay, OVO, or DANA are unavailable for both.

The quick rule of thumb: $20 for Pro equals approximately 460 steps Claude Sonnet 5 through an API key in Kunavo, or approximately 23 tasks containing 20 steps. If you open Claude Code almost every working day and rarely hit the limit, Pro is cheaper. If you use it only when you have a project, an API key is more economical because weeks without coding are not billed.

Sources, as of October 3, 2026: claude.com/pricing, Anthropic Help Center articles What is the Pro plan?, Paid plan billing FAQs, and What is the Max plan?. Kunavo prices are from the model catalog (as of October 3, 2026). All figures are in US dollars because both providers publish prices in dollars.

Is Claude Code free?

Only the tools are free: the CLI in the terminal and the editor extension. Every request to a model, including those sent by subagents behind the scenes, still has to be paid for by someone. According to the Claude Code installation documentation (checked on October 3, 2026), Claude Code requires a Pro, Max, Team, Enterprise, or Console account; a free claude.ai account is not enough.

What about “free API credits”? According to the Anthropic API pricing page (checked on October 3, 2026), new users do receive a small amount of credit to try the API. After that, API credits purchased directly from Anthropic are prepaid, expire one year after purchase, and are non-refundable (Anthropic Help Center, checked on October 3, 2026).

If your priority is the lowest initial cost, choose a prepaid key. In Kunavo, you can top up starting at $10, and the balance never expires. $10 is equivalent to approximately 230 agent steps on Claude Sonnet 5, or roughly 12 small tasks containing 20 steps, calculated without cache.

Subscription vs. API key: the difference in one table

One variable determines who is billed. As long as ANTHROPIC_AUTH_TOKEN or ANTHROPIC_API_KEY is set in the Claude Code environment, each request is paid for using that key, not the subscription; none of your plan limits are used. The article Use Claude Code with your Pro or Max plan explains this for ANTHROPIC_API_KEY, and the LLM gateway documentation covers both (checked on October 3, 2026). Your claude.ai login is not removed; it is simply not used. When you remove the variable, Claude Code returns to the subscription (gateway connection documentation). The /status command shows which path is currently active.

Claude Code subscription pricing: Pro, Max, and Team

Indonesia is on Anthropic's list of supported countries for both Claude.ai and the API. Plan prices are in US dollars before tax, from claude.com/pricing and the Max plan article (checked on October 3, 2026):

Two small notes from the same source. The Max price above is the web subscription price; Anthropic says the price in mobile apps may differ. And in the feature table at claude.com/pricing, model access for Pro is listed as “Usage credits”, meaning it is paid separately at API rates.

Why does the price of Claude Pro in rupiah vary between articles?

Because Anthropic does not publish them. claude.com/pricing lists only dollar prices and says they exclude tax. The What is the Pro plan? article states that the US price is $20 per month, that local currency prices are available “where supported,” and that in some regions tax is included in the displayed price while in others it is added at checkout. The article does not say whether the Indonesian rupiah is a supported currency.

As a result, each author calculates it differently: some use that day's exchange rate, some add an estimated VAT, and some quote the App Store price. This page deliberately does not list rupiah amounts or VAT rates; the binding amount is the total shown at Anthropic checkout, or in the App Store and Google Play if you subscribe through the app.

Can you pay for a Claude subscription with QRIS or an e-wallet?

Not on the web. The Paid Plan Billing FAQs (Indonesian version, checked on October 3, 2026) state that Anthropic “only accepts credit or debit cards” for subscriptions purchased on the web. QRIS, GoPay, OVO, DANA, and ShopeePay are not included. Subscriptions through the Claude app on iPhone or Android are billed by Apple or Google, and payment options are set by each app store, not Anthropic. If your Visa or Mastercard is declined on the web, first check whether international online transactions are enabled in your banking app.

Five-hour and weekly limits

What you are actually buying with Pro or Max is quota. That quota is measured per session, reset every five hours, with an additional weekly limit for all models (Pro plan article), and is shared by Claude chat and Claude Code (Claude Code article for Pro or Max). Anthropic does not publish the exact amounts. A longer explanation of limit windows is available in the Claude limits guide (in English).

When you reach your limit in the middle of a task, you can wait for it to reset, upgrade your plan, or enable usage credits. Usage credits are prepaid at standard API rates (Anthropic Help Center), and while using them, the cache lasts only five minutes, rather than one hour as it does with a subscription, unless you set a one-hour TTL yourself (cost documentation; both checked on October 3, 2026). So if you choose usage credits, work beyond your limit is billed per token. At that point, it is worth comparing with using your own key: on Kunavo, every Claude model in the API table below costs less than Anthropic's list price.

Claude Code pricing with an API key

With a key, Claude Code is billed just like any other application that calls the API: per token, with no plan and no usage limit window. Files read and tool outputs count as input; code written, explanations, and thinking tokens count as output. The following table lists prices in dollars per 1 million tokens (input / output). List prices are from the Anthropic API pricing page (checked on October 3, 2026); Kunavo prices are from the catalog (as of October 3, 2026):

Still seeing Claude Sonnet 5 at $3/$15? Those are old figures. Anthropic briefly announced $2/$10 as introductory prices through August 31, 2026, with an increase to $3/$15 starting September 1. That increase was canceled, and the introductory prices became the standard prices (Anthropic API pricing page, checked October 3, 2026).

“One step” on this page means one round trip with the model within a task. Sent: the system prompt, conversation history, and relevant file excerpts, approximately 25,000 tokens. Returned: an edit or explanation, approximately 1,200 tokens. The per-step figures above are calculated without cache; how much the cache saves is shown in the next section.

Prompt cache: the largest cost component

On every step, Claude Code resends the entire conversation to the model. Most of the input is therefore the same context repeated, and that portion is billed much more cheaply when read from the prompt cache. Kunavo cache rates (per 1 million tokens, 5-minute window; the multiplier is calculated from that model's input price):

You notice the difference after a lunch break or meeting. With an API key, the cache lasts five minutes by default, compared with one hour on a subscription (cost documentation, checked on October 3, 2026). When the cache expires, the next message writes the entire context to the cache again. For 200,000 context tokens in Claude Sonnet 5, reloading costs about $0.35 on Kunavo, while a warm cache would cost only $0.028. If you switch to another task after a break, run /clear so you do not keep paying for the old context. Details of how this works are in the caching documentation (in English).

Claude Code cost simulation for one session

Rather than make up figures, we use an official example. The Claude Code costs page (checked October 3, 2026) shows the output /usage for one Sonnet 4.6 session: 1,200 input tokens, 5,300 output tokens, 940,000 read from cache, and 50,000 written to cache, worth $0.55 at list rates. We recalculated the same token counts at the rates for four models. We ignore tokenizer differences between models, price cache writes at the 5-minute window rate (the default for API keys), and exclude background usage; according to the same documentation, that is usually less than $0.04 per session. Thinking is billed as output, so models that think for longer will push the figures up.

For Claude Sonnet 5: $0.258 on Kunavo, $0.368 at list rates. Approximately 51% of that Kunavo figure comes from cache reads; without caching, the same session would be much more expensive. With the Pro subscription money ($20), you could pay for approximately 77 sessions of this size through Kunavo. For comparison, the same session on Claude Opus 5.5 costs approximately $0.384, on Claude Opus 5 approximately $0.645, and on Claude Haiku 4.5 approximately $0.129.

These figures are arithmetic, not measured charges, and your session may be much smaller or larger. The most honest approach: open /usage after a normal work session, copy its four figures into the following script, and see for yourself.

# Biaya satu sesi Claude Code dari jumlah tokennya sendiri (USD, tarif Kunavo).
# Perintah /usage di Claude Code menampilkan empat angka untuk sesi yang berjalan.
TARIF = {  # per 1 juta token: (input, output, baca cache, tulis cache)
    "claude-sonnet-5": (1.40, 7.00, 0.14, 1.75),
    "claude-opus-5-5": (2.80, 14.00, 0.14, 3.50),
    "claude-opus-5": (3.50, 17.50, 0.35, 4.375),
    "claude-haiku-4-5": (0.70, 3.50, 0.07, 0.875),
}

def biaya_sesi(model, token_input, token_output, baca_cache, tulis_cache):
    i, o, r, w = TARIF[model]
    return (token_input * i + token_output * o + baca_cache * r + tulis_cache * w) / 1e6

# Contoh sesi dari dokumentasi biaya Anthropic:
print(round(biaya_sesi("claude-sonnet-5", 1_200, 5_300, 940_000, 50_000), 3))  # 0.258

The dollar amount shown in /usage is Claude Code’s own estimate at Anthropic list rates (cost documentation). If you use Kunavo, the actual balance deduction is recorded in your dashboard’s usage history. For other scenarios, try the Claude token cost calculator.

For an enterprise-scale picture, the cost documentation gives averages of about $13 per developer per active day, $150–250 per developer per month, and 90% of users below $30 per active day. These are enterprise deployment figures at Anthropic rates, not Kunavo rates.

Break-even point: when Pro, Max, or an API key is cheaper

The formula is simple: plan price ÷ cost of one step = number of steps per month you must reach for the subscription to be more economical. Calculated using the pre-tax plan price and uncached steps:

Caching shifts that figure. If 20,000 of the 25,000 input tokens are read from the cache and the rest are written, one step drops to approximately $0.020, and $20 now equals approximately 1,000 steps. A month of intensive work (20 days × 100 steps) on Claude Sonnet 5 through Kunavo ranges from approximately $87 without caching to $40 with a warm cache. Also remember that Anthropic has not announced how many steps fit within a plan's quota, so this table only shows the break-even point, assuming the limit is sufficient.

Translated into everyday work patterns:

• Coding every workday, rarely hitting the limit: choose Pro. $20 through an API key covers only approximately 460 steps, or roughly 5 intensive workdays.
• Coding all day, almost every day: this is where Max 5x starts to make sense. Its break-even point is approximately 2,300 steps per month, or roughly 23 intensive workdays; above that, Max is cheaper than a key, as long as the quota is sufficient.
• Freelancers who are busy only when they have a project: API key. Slow months are not billed, and busy months are paid for according to the work performed.
• Small teams, CI, or automated scripts: API key, one key per person or per pipeline, so usage for each component remains visible separately.
• Already on Pro but often running out of quota: stay on Pro, and use an API key only during blocked hours, rather than buying usage credits at API rates.

Upgrade to Max: when is it worthwhile?

The difference between Max 5x and Pro is $80 per month. If you put that amount into Kunavo balance, it equals approximately 1,840 Claude Sonnet 5 steps without caching, or approximately 37 work afternoons containing 50 steps (about $2.17 each). So the question is: how many afternoons per month are you actually blocked by the Pro limit? If more than that, Max is cheaper. If fewer, it is more economical to stay on Pro and pay for those afternoons with an API key.

Installing a Kunavo API key in Claude Code

No plugin is required. Claude Code reads ANTHROPIC_BASE_URL, so simply point it to Kunavo, which speaks the Anthropic Messages API format. On macOS, Linux, or WSL, paste this line into the terminal for a single session, or into ~/.zshrc / ~/.bashrc to make it permanent:

export ANTHROPIC_BASE_URL=https://api.kunavo.com   # cukup domainnya, tanpa /v1
export ANTHROPIC_AUTH_TOKEN=sk-kn-...
export ANTHROPIC_MODEL=claude-sonnet-5
export ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-5-5
export ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-5
export ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-haiku-4-5

Windows users can first try it through PowerShell:

# Hanya berlaku untuk jendela PowerShell ini; tutup jendela, hilang.
$env:ANTHROPIC_BASE_URL = "https://api.kunavo.com"
$env:ANTHROPIC_AUTH_TOKEN = "sk-kn-..."
$env:ANTHROPIC_MODEL = "claude-sonnet-5"
$env:ANTHROPIC_DEFAULT_OPUS_MODEL = "claude-opus-5-5"
$env:ANTHROPIC_DEFAULT_SONNET_MODEL = "claude-sonnet-5"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL = "claude-haiku-4-5"
claude

Several things commonly cause setup failures:

• ANTHROPIC_BASE_URL should contain only the domain, without /v1. Claude Code adds the /v1/messages path itself.
• Use ANTHROPIC_AUTH_TOKEN. This variable is sent as the Authorization: Bearer header, while ANTHROPIC_API_KEY is sent as x-api-key (gateway connection documentation, checked October 3, 2026). Kunavo accepts both, but ANTHROPIC_API_KEY in interactive mode waits for approval once, and a key you have rejected there before will be ignored without asking again. ANTHROPIC_AUTH_TOKEN is used immediately.
• For long-term use, especially on Windows, save the variable in the env block of the user settings file (on Windows: %USERPROFILE%\.claude\settings.json). If the same variable also exists in the shell, the value from this file takes precedence.
• Do not store the key in the project's .claude/settings.json: that file will be committed to the repository. For the VS Code extension, enter the variable in claudeCode.environmentVariables in the VS Code user settings.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.kunavo.com",
    "ANTHROPIC_AUTH_TOKEN": "sk-kn-...",
    "ANTHROPIC_MODEL": "claude-sonnet-5",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-5-5",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-5",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-4-5"
  }
}

Why specify the model name explicitly? Model configuration documentation (checked October 3, 2026) says that in the Anthropic API, Claude Code’s default model and the opus alias point to Opus 5.5, while the sonnet alias points to Sonnet 5.5, and those aliases move to newer versions as they are released. Kunavo does not serve Sonnet 5.5; without ANTHROPIC_DEFAULT_SONNET_MODEL, /model sonnet, the opusplan execution phase, and subagents with model: sonnet will request that model and receive a 404. With ANTHROPIC_MODEL, a session starts on the less expensive Claude Sonnet 5; ANTHROPIC_DEFAULT_SONNET_MODEL also directs the sonnet alias to Claude Sonnet 5; ANTHROPIC_DEFAULT_HAIKU_MODEL also sets the model for background tasks; and the opus alias is directed to Claude Opus 5.5 ($2.80 / $14.00 on Kunavo), which requires Claude Code v2.1.280 or later (on older versions, run claude update). Thinking is always enabled on this model, and its tokens are billed as output (cost documentation).

Check the result: run claude, then type /status. The Anthropic base URL line should be https://api.kunavo.com and the Auth token line should mention ANTHROPIC_AUTH_TOKEN. If it instead shows Login method with a claude.ai account, the variable has not been read by that session. The key (starting with sk-kn-) is created on the API key page after you sign up. Complete integration guide (in English): Claude Code with Kunavo.

Add funds from Indonesia: Visa, Mastercard, Link, Google Pay

This section applies only to the API key route. Kunavo is an API gateway compatible with the Anthropic and OpenAI formats; you add balance through a Stripe Checkout page with prices in US dollars. Stripe Adaptive Pricing documentation (checked October 3, 2026) lists Indonesia among the markets where prices can be displayed in local currency. If your checkout displays rupiah, Stripe’s exchange rate already includes a 2–4% conversion fee paid by the buyer. Paying in dollars avoids that fee, but the card issuer may charge its own exchange rate and foreign transaction fee.

• Credit or debit cards (Visa, Mastercard). Debit cards are supported as long as they bear the Visa or Mastercard logo and have been enabled for international online transactions in the banking app.
• Link: Stripe's wallet that stores your card, so subsequent top-ups require only one click.
• Google Pay in Chrome or Android, if your card is already saved there. Indonesia is on the Google Wallet country list (checked October 3, 2026); see the Google Pay guide (in English) for the steps.
• Apple Pay appears only in Safari or on an iOS device where it is enabled, and Indonesia is not on the Apple Pay country list (checked October 3, 2026), so do not count on it.

Unavailable: QRIS, GoPay, OVO, DANA, ShopeePay, virtual accounts, and bank transfers. Kunavo also does not issue tax invoices or VAT documents; top-up records (date, amount, status) can be viewed in the Billing panel.

• Top-ups start at $10; larger amounts may receive a balance bonus: $100 → balance $110, $1,000 → balance $1,200, and $5,000 → balance $6,250.
• Prepaid, with no monthly fee, balance never expires, and failed requests are not charged.
• Automatic top-ups are available only for saved cards or Link.
• This balance is only for the Kunavo API. Claude Pro or Max subscriptions are still paid to Anthropic: on the web by card, and in the app through the App Store or Google Play.

A more complete list of Claude API rates is available in the Claude API pricing guide, and all model prices are on the Kunavo pricing page. To get started, create a Kunavo account, add funds, then create a key. English version of this page: Claude Code pricing.

What you need to know before switching to an API key

Claude Code’s core features (agents, tools, subagents, MCP servers) run on the Messages API and continue to work with a gateway key. Remote Control and voice dictation are different: they require a claude.ai login and are unavailable while a gateway key is active (gateway connection documentation, checked October 3, 2026).

Kunavo serves Claude from shared capacity without dedicated quotas; if you need guaranteed capacity, buy directly from Anthropic. Anthropic also states that it does not support, maintain, or audit third-party gateways (LLM gateway documentation). And if, after reading all this, it turns out you code every day without ever hitting a limit, Pro is probably still the cheaper option.

Claude Code pricing FAQ