Claude Code pricing 2026: Pro and Max subscription or API key? Cached session cost and card payment from Türkiye

Downloading and installing Claude Code is free; the cost comes from the Claude model called by each command. You cover this cost either with a fixed monthly subscription (Claude Pro, Max, Team) or an API key that charges you for the tokens you use. Claude Code cannot be enabled with a free Claude account.

• Subscription: Claude Pro monthly $20 (monthly $17 with annual payment, $200 upfront), Claude Max monthly $100 (5x) or $200 (20x). Anthropic's US prices are before tax; web subscriptions can be paid for by card only.
• API key: no monthly fee. At Kunavo, 1M input tokens for Claude Sonnet 5 cost $1.40, and 1M output tokens cost $7.00; an average agent step (approximately 25,000 input and 1,200 output tokens) costs approximately $0.0434 without caching.
• Break-even point: Pro's $20 price covers approximately 460 steps with a key. If Claude Code is open almost every day and the Pro limits are sufficient, choose the subscription; if you take weeks off, a key costs less.
• Payment: Top up your Kunavo balance with a Visa or Mastercard card or Link, starting at $10; the balance does not expire.

Source and date of the figures. Plan prices are from claude.com/pricing and the Anthropic Help Center articles on the Pro plan, Max plan and billing FAQs, read on October 3, 2026. Kunavo prices are from the model catalog, as of October 10, 2026. Both publish prices in dollars, so the amounts are in dollars; TRY and VAT are covered separately below.

Subscription: Claude Pro, Max, and Team prices

According to the setup documentation, using Claude Code with an Anthropic account requires a Pro, Max, Team, Enterprise or Console (API) account; a free claude.ai account is not enough. If you use a gateway key, you do not need this login at all. Türkiye is on Anthropic’s list of supported countries for both Claude.ai and the API. Plan prices in dollars (excluding tax, from claude.com/pricing and the Max article, October 3, 2026):

Two details: Anthropic states that the prices it gives for Max apply to web subscriptions, and that the price may differ for subscriptions purchased through the iOS or Android app. According to claude.com/pricing, Fable models work on Pro only through usage credits (additional usage paid at API prices); on Max, at most half of the weekly limit can be spent on Fable.

How many TRY does Claude Pro cost, and is VAT added to the price?

Anthropic does not publish prices in TRY for Türkiye and states that the figures on its pricing page exclude tax. The Pro plan article gives only the general rule: prices are shown in local currency in supported regions; monthly fees can vary by region; and tax is included in the price in some regions and added at checkout in others. It does not say whether TRY is one of those currencies.

We have two reliable pieces of information about tax. Türkiye’s standard VAT rate is 20% (PwC Worldwide Tax Summaries, read on October 3, 2026). According to Anthropic, your billing address determines which tax applies (tax ID article; Türkiye is not mentioned separately). Since none of the TRY prices circulating on Turkish-language websites cites a source, we do not give a TRY price here: the applicable amount is the one Anthropic shows you at checkout.

Paying for the subscription with a Turkish card

According to Anthropic’s billing FAQs, a subscription purchased on the web can be paid for only by credit or debit card; if you subscribed in the app, the App Store or Google Play handles the charge. To pay a foreign merchant with a card from a Turkish bank, the card must be enabled for international use; the bank may also charge an international transaction fee and a foreign currency conversion fee, with rates listed in the card’s fee schedule (HesapKurdu guide, read on October 3, 2026).

You cannot pay for Pro or Max through Kunavo. The card and Link payments described on this page go only toward the API balance on Kunavo; Anthropic itself collects payment for the Claude subscription.

The key subscription metric: limits

A subscription does not buy unlimited usage. Pro and Max usage is measured by two counters: a session limit that resets every five hours and a weekly limit that covers all models (Pro and Max plan articles). Chatting on Claude.ai and working in Claude Code count against the same limits (Claude Code with a Pro/Max plan article). The English guide Claude Pro and Max limits explains in detail how the windows work.

When a limit is reached, you have three options: wait for it to reset, upgrade your plan or enable usage credits. Usage credits are prepaid and billed at standard API rates (Anthropic Help Center); in this mode, the cache is retained for five minutes rather than one hour (cost documentation; both read on October 3, 2026). That means every token is already billed at API rates once you exceed the limit. When you do the same work with your own key, only the price changes: Claude models on Kunavo are 30% below Anthropic’s list price.

Is there a free option?

No. The CLI and editor extensions are free, but every request goes to a model, and model calls are billed, including those made by subagents. The Free plan does not include Claude Code; the Anthropic API also counts tokens from the first request. API credits purchased in advance through Anthropic Console expire one year after purchase and are nonrefundable (Anthropic Help Center, read on October 3, 2026).

The no-commitment option is a prepaid balance: at Kunavo, you load at least $10, there is no monthly fee, and the balance does not expire. Without accounting for caching, this amount covers approximately 230 agent steps with Claude Sonnet 5, or approximately 12 jobs consisting of 20 steps.

Claude Code with an API key: token prices by model

With a key, Claude Code is billed like a regular API client: there is no limit window, and each request is charged according to its token count. Files Claude Code reads, tool outputs and conversation history count as input; code, explanations and thinking tokens it generates count as output. The table lists dollar prices per 1M tokens (input / output). The Anthropic column is from claude.com/pricing (October 3, 2026); the Kunavo column is from the catalog (as of October 10, 2026):

In this guide, “agent step” means one request Claude Code sends to the model during a task and the response it receives. The input contains the system prompt, the conversation so far, and newly read files; the output contains an edit or a short explanation. Without accounting for caching, that is roughly 25,000 input and 1,200 output tokens. The English version of the guide, Claude Code pricing, uses the same unit, so you can compare the results side by side.

Most of the bill comes from the cache

Claude Code resends the entire conversation with every request. As a result, most of the input is context the model has seen before, and prompt caching reads it much more cheaply. At Kunavo, cache prices (per 1M tokens, 5-minute window; the multiplier is applied to the model's input price) are:

Taking a break has a cost too. According to the cost documentation (October 3, 2026), the cache is retained for five minutes by default with an API key and for one hour with a subscription. After a break of more than five minutes, your first message does not read the context from the cache; it writes it again from scratch. With a context of Claude Sonnet 5 and 200,000 tokens, that costs about $0.35 in Kunavo for a single pass; reading the same context while the cache is warm costs about $0.028. If you are switching to another task after a break, clear the context first with /clear. Details: cache documentation (in English).

The actual cost of a session

Anthropic publishes an example /usage output in its cost documentation (read on October 3, 2026). We took the token counts from that output and recalculated the cost using Kunavo prices. The assumptions for the calculation are:

• The figures are as in the documentation: 1,200 input, 5,300 output, 940,000 cache read, and 50,000 cache write. In the documentation, this is a Sonnet 4.6 session and costs $0.55 at list price.
• Token counts are fixed; only the model price changes. Differences between the models' tokenizers are disregarded.
• Cache writing uses the price for the 5-minute window because that is the default window for an API key.
• Thinking tokens are included in the output (as in the same documentation); the longer the model thinks, the larger the output share becomes.
• Claude Code's background calls are not included in the calculation; according to the documentation, they are generally under $0.04 per session.

With Claude Sonnet 5, this session costs approximately $0.258 on Kunavo and approximately $0.368 at list price. Approximately %51 of the amount goes to cache reads: the cache is what primarily determines the bill. At Pro's monthly price of $20, you could run approximately 77 sessions of this size on Kunavo. The same session costs approximately $0.384 with Claude Opus 5.5, approximately $0.645 with Claude Opus 5, and approximately $0.129 with Claude Haiku 4.5.

Read this as an estimate, not as measured spending or an upper limit. To price your own session, enter /usage in Claude Code and put the four token counts it displays into the following script:

# Bir Claude Code oturumunun maliyeti, kendi token sayılarından (USD, Kunavo fiyatları).
# Claude Code'da /usage, geçerli oturumun dört token sayısını gösterir.
ORANLAR = {  # 1M token başına: (girdi, çıktı, önbellekten okuma, önbelleğe yazma)
    "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 oturum_maliyeti(model, girdi, cikti, onbellek_okuma, onbellek_yazma):
    i, o, r, w = ORANLAR[model]
    return (girdi * i + cikti * o + onbellek_okuma * r + onbellek_yazma * w) / 1e6

# Anthropic maliyet belgelerindeki örnek oturum:
print(round(oturum_maliyeti("claude-sonnet-5", 1_200, 5_300, 940_000, 50_000), 3))  # 0.258

The dollar figure on the /usage screen is Claude Code's own estimate using Anthropic's list prices; the usage history in the panel shows the amount actually deducted from your Kunavo balance. For other scenarios, you can use the Claude token cost calculator.

For comparison, Anthropic's enterprise figures: according to the cost documentation, the average is approximately $13 per developer per active day and $150–250 per developer per month; 90% of users stay below $30 on an active day. These are Anthropic's enterprise averages, not Kunavo pricing.

Pro, Max, or a key? Break-even calculation

You can reduce the comparison to a single number: plan price ÷ the cost of one agent step. The resulting number is the monthly step count at which the subscription starts to be cheaper than a key. Using plan prices before tax and no caching:

When the cache is warm, the calculation shifts in favor of the key: when 20,000 of 25,000 input tokens are read from the cache and the rest are written, one step costs approximately $0.020, and $20 buys approximately 1,000 steps. A busy month (20 workdays, 100 steps per day) costs approximately $87 uncached at Kunavo with Claude Sonnet 5, and approximately $40 with a warm cache. The table shows only the point after which the fixed fee remains cheaper; whether that number of steps fits within the plan limits is a separate question.

Is upgrading from Pro to Max 5x worth it?

The difference is $80 per month. With the same money, at Kunavo you get approximately 1,840 uncached steps with Claude Sonnet 5. To make it concrete, count an afternoon when you hit the limit as 50 steps: the difference corresponds to approximately 37 afternoons, each costing approximately $2.17 with a key. If you hit the Pro limit on more afternoons than this per month, Max makes sense; if less often, staying on Pro and switching to a key only during those hours is cheaper.

The subscription is disabled when a key is configured

Claude Code does not use both payment methods at the same time. If ANTHROPIC_AUTH_TOKEN is present in the environment (or if you approved ANTHROPIC_API_KEY once in interactive mode), Claude Code sets aside your subscription login and uses the key: plan limits no longer apply, and the key’s owner is billed for the tokens (LLM gateway documentation and the Claude Code with a Pro/Max plan article, October 3, 2026). Remove the variable to return to the subscription; /status tells you which method is active at that moment.

Setup: connect Claude Code to a Kunavo key

Claude Code can be directed to an Anthropic Messages API-compatible endpoint with ANTHROPIC_BASE_URL; no separate plugin is required for Kunavo. To test it, set the variables in the terminal; if you want them to persist, add them to ~/.zshrc:

export ANTHROPIC_BASE_URL=https://api.kunavo.com   # yalnızca adres, sonuna /v1 eklemeyin
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

Use only the domain name in the address; Claude Code adds the /v1/messages path itself. We recommend using ANTHROPIC_AUTH_TOKEN for the key. According to the gateway connection documentation (October 3, 2026), this variable is sent with the Authorization: Bearer header, while ANTHROPIC_API_KEY is sent with the x-api-key header. Kunavo accepts both headers; the difference is on the Claude Code side: ANTHROPIC_AUTH_TOKEN takes effect immediately, while ANTHROPIC_API_KEY requires one approval in interactive mode. The env block in user settings is a more reliable place than a shell file; if the same variable is defined in both places, the setting in the configuration file takes precedence:

{
  "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"
  }
}

Do not put the key in the project's .claude/settings.json file; that file is committed to the repository. In the VS Code extension, variables are entered under claudeCode.environmentVariables in VS Code user settings.

Why are the model variables written out explicitly? According to the model configuration documentation (October 3, 2026), Claude Code uses Opus 5.5 by default with the Anthropic API; the opus alias maps to Opus 5.5, and the sonnet alias maps to Sonnet 5.5, with aliases changing as new versions are released. Kunavo does not offer Sonnet 5.5: while ANTHROPIC_DEFAULT_SONNET_MODEL is not defined, selecting sonnet in the /model menu, opusplan entering its execution phase, or sonnet written in a subagent definition sends the request to a model with no Kunavo equivalent, which returns a 404 error. ANTHROPIC_MODEL starts a session with the less expensive Claude Sonnet 5 model, ANTHROPIC_DEFAULT_SONNET_MODEL maps the sonnet alias to the same model, and ANTHROPIC_DEFAULT_HAIKU_MODEL pins the model for background tasks. ANTHROPIC_DEFAULT_OPUS_MODEL maps the opus alias (including the planning phase of opusplan) to the Claude Opus 5.5 model ($2.80 / $14.00). This model requires Claude Code v2.1.280 or later; if your version is older, run claude update first. Thinking cannot be disabled for the Claude Opus 5.5 model, and thinking tokens are billed as output (cost documentation).

For verification, open the claude command and enter /status: you should see https://api.kunavo.com on the Anthropic base URL line and ANTHROPIC_AUTH_TOKEN on the Auth token line. If you do not, the variable did not reach this session. A key beginning with sk-kn- is created in the key panel after you sign up. Detailed setup (English): Claude Code with Kunavo.

Top up your Kunavo balance from Turkey: cards and Link

Kunavo is a gateway compatible with the Anthropic and OpenAI APIs; you can add funds for Claude Code usage through Stripe’s payment page. Prices are in dollars. Stripe determines whether the amount appears in TRY or dollars at checkout: Türkiye is among the supported markets in Stripe’s Adaptive Pricing documentation, so the amount may be shown in TRY. In that case, the displayed exchange rate includes a 2–4% conversion fee paid by the buyer; if you choose to pay in dollars, that fee does not apply, but your bank’s own exchange rate and international transaction fee may apply instead (read on October 3, 2026). Payment methods available from Türkiye:

• Card (Visa, Mastercard): the card must be enabled for international use; this setting is enabled through mobile banking or online banking (HesapKurdu guide). If payment is declined, this setting is the first thing to check.
• Link: Stripe's saved payment service; pay with the card you saved in it.

As for Apple Pay and Google Pay: Kunavo’s payment page shows these wallets on devices and accounts that support them, but Türkiye is listed neither in Apple’s Apple Pay country list nor in Google’s Google Pay and Google Wallet lists (checked on October 3, 2026). In practice, from Türkiye you can pay by card or Link.

Unavailable: installments, Papara, bank transfer/EFT and PayPal. Cards that operate only on the Troy network are also not accepted; Troy is not among the card brands Stripe supports. Kunavo does not issue VAT invoices, e-Fatura or e-Arşiv invoices for any country; a record of top-ups (date, amount, status) is available in the Billing panel.

• Minimum top-up $10; larger top-ups receive a balance bonus: $100 → $110 balance; $1.000 → $1.200 balance; $5.000 → $6.250 balance.
• The balance is prepaid; there is no monthly fee, it does not expire, and failed requests are not charged.
• Automatic top-up works only with a saved card or Link. Each top-up is a single charge.
• These methods only top up the API balance on Kunavo; they cannot be used to pay for Claude Pro or Max, and Anthropic collects subscription payments itself on the web by card only.

For API prices by model and payment from Turkey, see also: Claude API pricing. To get started, create an account on Kunavo, top up your balance, and create a key. Prices for all models are on the pricing page.

What you give up with a key

The agent loop, tools, subagents and MCP servers run through the Messages API, so they work the same way with a gateway key. Two features are tied to a claude.ai account and are disabled while the gateway key is active: Remote Control and voice dictation (gateway connection documentation, October 3, 2026). To turn them back on, sign in with claude.ai and remove the key variable; for Remote Control, you must also delete the ANTHROPIC_BASE_URL variable—according to the documentation, signing in alone is not enough.

Kunavo offers Claude on shared capacity; there is no capacity reserved for you and no contractual guarantee. If you need guaranteed capacity, work directly with Anthropic. Anthropic also explicitly says it does not endorse, maintain or audit third-party gateways (LLM gateway documentation). In short: if you code every day and stay within Pro limits, Pro is probably cheaper. A key is suited to people who work intermittently, teams with per-person budgets, CI and times when the limit has run out.

Frequently asked questions