Claude API pricing 2026: pricing by model, cost per request, and how to pay from Vietnam

How much does the Claude API cost? Short answer: there is no monthly price. You pay for the number of tokens used, and each model has two rates, one for the tokens you send (input) and one for the tokens the model generates (output), calculated in USD per 1 million tokens. The figures to know first (prices checked October 3, 2026):

• Claude Sonnet 5, the first model to try: $2.00 / $10.00 at Anthropic, $1.40 / $7.00 on Kunavo.
• The cheapest is Claude Haiku 4.5 ($0.70 / $3.50 on Kunavo); the most advanced and expensive is Claude Fable 5.1 ($7.00 / $35.00). Every model in the table is 30% below Anthropic's price.
• A coding agent step on Sonnet 5 costs about $0.0434, and only about $0.0167 with caching.
• Top up with an international Visa/Mastercard (debit cards are also accepted), Link, or Apple Pay / Google Pay if your device supports them; minimum $10, no monthly fee, and the balance never expires.

Anthropic's prices are from the API pricing documentation and claude.com/pricing, checked on October 3, 2026. Kunavo's prices are not entered manually in the article: the page pulls them directly from the model catalog the system uses to calculate charges, so the amount you see is also the amount deducted from your balance.

Claude API pricing 2026

USD per 1 million tokens, listed in input / output order. Kunavo's reduction is the same in both directions.

The “Cache reads” column is the price for repeated context when prompt caching is enabled, calculated as a percentage of the input price; the cache section below explains why this column matters for agents. Three things the pricing table does not say by itself:

• Long prompts do not cost extra. Except for Claude Haiku 4.5 (a 200,000 token window), all models in the table have a 1 million token context window, and according to Anthropic (as of October 3, 2026), a 900,000-token request costs the same unit rate as a 9,000-token request. Kunavo works the same way.
• The same text produces more tokens with a newer model. Anthropic says models from Claude 4.7 onward (including Sonnet 5, Opus 5, Opus 5.5, and Fable 5.1) use a new tokenizer that produces about 30% more tokens for the same text, and token counts also vary by language (pricing documentation, as of October 3, 2026). Vietnamese and English prompts may therefore have different costs; measure with your actual text.
• Batch API is not included in the table. Anthropic reduces both input and output by 50% for requests processed asynchronously through the Batch API; the table above shows the price for standard requests.

GPT and image- and video-generation model prices are on the Pricing page; Sonnet 5's technical specifications are on the Claude Sonnet 5 page; the more detailed English version of this table is Claude API pricing 2026.

Putting it into dollars: one question, one agent step, one month

To find out how much you will spend, calculate based on the unit you actually call through the API. The table below uses two common scenarios, with caching disabled:

• A short chat question: 2,000 input tokens, 500 output tokens.
• One coding-agent step such as Claude Code: about 25,000 input tokens (system prompt, conversation history, code being edited, tool execution results) and about 1,200 output tokens.

Putting it into monthly terms: if an agent runs 200 steps per day on Sonnet 5 with caching enabled (the cache ratio shown in the next section), then 22 working days will cost about $73.61 on Kunavo. This is only an illustrative figure; to calculate based on your own usage, use the Claude token cost calculator.

One way to reduce costs: output tokens cost 5 times as much as input tokens, so long answers are what cost money, not long questions. Asking the model to answer concisely and setting max_tokens appropriately is usually more economical than shortening the prompt.

Prompt caching: why agents cost less than you think

Agents such as Claude Code resend nearly the entire conversation at every step. Content sent in the previous step, if still in the cache, is charged at only 10% of the input price (with this caveat: Claude Fable 5.1 is 2.5% and Claude Opus 5.5 is 5%). In return, tokens written to the cache for the first time (the 5-minute cache tier) are charged at 1.25× of the input price: pay a little extra on the first step so later steps cost much less. These multipliers are identical to Anthropic's (as of October 3, 2026). See the caching documentation for how to enable caching in a request.

Take the agent step above again (25,000 input tokens, 1,200 output tokens), assuming 22,000 input tokens are read from the cache and the remaining 3,000 tokens are new content written to the cache. The longer the session, the larger the proportion read from the cache usually becomes.

In other words, simply multiplying “number of tokens × rate” produces a figure much higher than what an agent actually spends.

Sonnet, Opus, Haiku, or Fable: choose by task

Sonnet 5: the starting point for most projects

For programming, agents, chatbots, and most applications: start with Claude Sonnet 5 ($1.40 / $7.00 on Kunavo), and switch only when you have a reason. Many Vietnamese articles still say that Anthropic's $2.00 / $10.00 pricing is promotional, “until the end of August, then rising to $3 / $15.” That information is outdated: according to the pricing page (as of October 3, 2026), the planned increase on September 1, 2026 will not take place, and these are the official prices.

Still using Claude Sonnet 4.6? It costs more ($3.00 / $15.00 at Anthropic, $2.10 / $10.50 on Kunavo), so unless you need to preserve the old behavior, switching to Sonnet 5 saves money immediately. On claude.com/pricing (as of October 3, 2026), Anthropic has introduced Claude Sonnet 5.5 as its latest Sonnet, at the same price as Sonnet 5, and placed Sonnet 5 in the “Legacy” category even though the API still serves it. Kunavo currently offers Sonnet 5, not Sonnet 5.5.

Opus 5.5 or Opus 5: the newer version is cheaper again

When Sonnet is not powerful enough (long tasks, many steps, complex code), move up to Claude Opus 5.5. At Anthropic it costs $4.00 / $20.00, lower than the previous generation, Claude Opus 5 ($5.00 / $25.00); on Kunavo it is $2.80 / $14.00 compared with $3.50 / $17.50. Opus 5.5 is also cheaper for cache reads: 5% of the input price, while Opus 5 is 10%. Opus 5 is worth keeping only if you already have a custom-tuned prompt set for it. And if the task does not truly require Opus, Sonnet 5 is cheaper than both.

Haiku 4.5 and Fable 5.1: the two ends of the pricing table

Claude Haiku 4.5 ($0.70 / $3.50 on Kunavo) suits high-volume work that does not require deep reasoning: classification, data extraction, short summaries, and simple questions. At the other end, Claude Fable 5.1 is the most advanced and expensive model in the table ($10.00 / $50.00 at Anthropic, $7.00 / $35.00 on Kunavo); use it only for problems that other models have already tried and failed to solve.

Paying from Vietnam: Visa/Mastercard, Link, Apple Pay, Google Pay

If you buy directly from Anthropic: Anthropic charges only in USD, and standard accounts pay with major credit cards; new accounts receive a small amount of credit to try it (pricing page, as of October 3, 2026). Credit purchased in the Anthropic Console is prepaid, expires one year after purchase, and is non-refundable (Anthropic Help Center, as of October 3, 2026). Vietnam is on Anthropic's list of supported countries for both the API and Claude.ai (as of October 3, 2026), so articles claiming that Vietnam “is not supported” are incorrect.

Kunavo accepts payments through Stripe. When topping up from Vietnam, you can choose:

• an international Visa or Mastercard; both international credit and debit cards are accepted (the payment page also accepts Amex, JCB, and UnionPay);
• Link, meaning a card you have saved with Stripe;
• Apple Pay or Google Pay, if set up on your device and browser. Both Apple Pay and Google Wallet are available in Vietnam (as of October 3, 2026).

MoMo, ZaloPay, VNPay, and bank transfers are not supported. Domestic ATM cards that carry only the Napas logo also cannot be used, because Napas is not on Stripe's list of accepted card brands (as of October 3, 2026). You need a card with a Visa or Mastercard logo.

Step by step

1. Create an account at kunavo.com using your email or Google account. You do not need to enter a card when signing up.
2. Go to the Billing page and choose the top-up amount. Minimum $10; larger top-ups receive a bonus (top-up amount → balance received: $100 → $110, $1,000 → $1,200, $5,000 → $6,250).
3. On the Stripe checkout page, check the amount and currency: the base price is in USD, but the page may display it in VND. See the fee section below for which currency you should use.
4. Choose a card or wallet, then authenticate the transaction with an OTP or 3-D Secure in your banking app if your bank requires it.
5. The balance is usually added after a few seconds. If it does not appear, reload the Billing page instead of paying again.
6. Go to the API Keys page to create a key. The key starts with sk-kn- and is displayed only once, so save it immediately.

If your card is declined, check in your banking app whether online and international payments are enabled, then try again.

Fees and notes

• Prices are listed in USD. If you pay in VND, Stripe's exchange rate includes a 2–4% conversion fee paid by the buyer, and the rate is locked for 24 hours (Stripe Adaptive Pricing, as of October 3, 2026). Some banks charge an additional fee of about 1.1% for VND transactions with merchants whose country code is outside Vietnam (the bank checks the country code in the payment request, not your current location); Kunavo's Stripe account is not in Vietnam, so this top-up is subject to the fee. Examples include Techcombank (Visa payment cards, 1.1% including VAT, effective August 13, 2023) and VPBank (international debit cards, 1.1% excluding VAT, effective November 15, 2023); both documents were reviewed on October 3, 2026.
• If you pay in USD, Stripe does not charge a conversion fee, but your bank may apply its own exchange rate and foreign transaction fee (Stripe, as of October 3, 2026); rates vary by bank. Check your card issuer's fee schedule before choosing.
• The balance never expires, there is no monthly fee, and failed requests are not charged.
• Automatic top-ups are available only with a saved card or Link.
• Kunavo does not issue Vietnamese VAT documents or VAT documents from any other country. On the Billing page, “Recent transactions” shows the date, amount, and status of each top-up; accounting questions can be sent to contact@kunavo.com.
• Top-up funds go only to your Kunavo API balance and cannot pay for Claude Pro or Max.

Running Claude Code with an API balance

Do not want to buy Pro or Max just to use Claude Code? You can run it with a Kunavo API key. Add the following six lines to ~/.zshrc (or ~/.bashrc), open a new terminal, and type claude:

export ANTHROPIC_BASE_URL=https://api.kunavo.com
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

The Claude Code documentation (as of October 3, 2026) says that once gateway credentials are available, Claude Code uses them instead of your claude.ai account, so subscription usage limits no longer apply. In return, each step deducts real funds from your balance at the rates in the table above: on Sonnet 5, about $0.0434 per step without cache, or about $0.0167 with cache. Note that ANTHROPIC_BASE_URL is only the base URL; do not append /v1.

Why pin the model: according to the Claude Code documentation (as of October 3, 2026), when running on Anthropic's API, Claude Code uses Opus 5.5 by default (more expensive than Sonnet 5), while the sonnet alias points to Sonnet 5.5, which Kunavo does not offer. ANTHROPIC_MODEL keeps the primary model on Sonnet 5. ANTHROPIC_DEFAULT_SONNET_MODEL maps the sonnet alias to Sonnet 5: without this line, /model sonnet, the execution phase of opusplan, and subagents configured with model: sonnet all return a 404 error. ANTHROPIC_DEFAULT_OPUS_MODEL pins the opus alias to Claude Opus 5.5 (claude-opus-5-5); this requires Claude Code v2.1.280 or later. On an older version, run claude update. ANTHROPIC_DEFAULT_HAIKU_MODEL controls the haiku alias and also selects the model for background tasks.

Kunavo accepts keys in both the Authorization: Bearer header and the x-api-key header. We recommend ANTHROPIC_AUTH_TOKEN (sent as Authorization: Bearer) because it takes effect immediately; with ANTHROPIC_API_KEY (sent as x-api-key), Claude Code asks you to confirm once before using the key in interactive mode (Claude Code documentation, as of October 3, 2026). Full setup instructions (in English): Claude Code with Kunavo. In Vietnamese: the Claude Code installation guide and the guide comparing subscriptions with API pricing, How much does Claude Code cost?.

Claude Pro, Max, and API: three different things

Many people search for “Claude pricing” but are actually asking about the chat subscription. The chat subscription and API are sold separately and billed differently (subscription prices are on claude.com/pricing, as of October 3, 2026; Anthropic specifies that listed prices exclude applicable taxes):

Kunavo top-up funds, whether paid by card or Link, go only to the API balance and cannot pay for Pro or Max. If you mainly chat in a browser, a subscription is the simpler choice; if you build an application or agent and want to pay as you go, choose the API.

Try Sonnet 5 now: create a Kunavo account, top up $10 with a Visa/Mastercard or Link, create an API key, and send your first request.

Frequently asked questions