Claude Prices in 2026: Pro, Max, the Per-Token API, and Which Is Cheaper

Claude has three pricing options, depending on how you use it. The app (claude.ai) is available by subscription: the Free plan costs $0; Claude Pro costs $20 per month ($17 per month with annual billing); and Claude Max starts at $100 per month, as of September 2026. The API has no monthly fee and is billed by token: on Kunavo, per million input / output tokens, Claude Haiku 4.5 costs $0.70 / $3.50; Claude Sonnet 5, $1.40 / $7.00; Claude Opus 5.5, $2.80 / $14.00; Claude Opus 5, $3.50 / $17.50; and Claude Fable 5, $7.00 / $35.00. Depending on the model, that’s between 30% and 30% below Anthropic’s official rate. Claude Code has no separate price: you pay through one of the two options above. All figures are in US dollars (USD).

A plan comparison doesn’t answer the question that decides whether to buy: at what level of usage does a subscription cost less than paying per token, and below what level does the reverse apply? This guide separates the three pricing options, gives current figures, and works through the calculation with numbers you can verify.

Rates verified on October 3, 2026. Plan prices are from claude.com/pricing (as of September 2026), and official API rates are from Anthropic’s pricing documentation. Kunavo rates are read from the site’s model catalog, so they match what you’re charged.

How much does Claude cost? The three options at a glance

1. Claude plans: Free, Pro, Max, and Team

A claude.ai subscription bundles chat, Claude Code, and app features into a fixed fee with usage limits. Limits are measured in rolling 5-hour windows and subject to a weekly cap. Chat and Claude Code draw from the same allowance. Prices as of September 2026, according to claude.com/pricing:

Plan prices change often; check the official page before subscribing.

Claude Pro price: $20 per month, and when to upgrade to Max

Claude Pro costs $20 per month with monthly billing, or $17 per month with annual billing ($200 upfront). More important than the $3 difference is that annual billing commits you for twelve months. It makes sense to pay for one or two months at a time, see how often you hit the 5-hour window limit, and decide afterward.

How often you hit the limit determines when to upgrade to Max (from $100 per month). If it stops you several times a week, the $80 difference pays for itself in time saved waiting. If it happens once a month, it’s cheaper to cover that peak with an API key. See Claude plan limits (in English) for your options when you run out of usage.

Is Claude free? What the Free plan includes

You can use Claude chat for free with the Free plan, which allows a limited number of messages per 5-hour window. The limit may vary with demand. Claude Code isn’t included in the Free plan.

The API has no permanent free tier: every call is charged. In return, there’s no monthly fee, so a month with no usage costs $0. In practice, to try Claude for free, use chat on the Free plan; alternatively, top up the $10 minimum and spend only on the calls you make.

2. Claude API pricing by model

The API is billed by token: one rate for input tokens (what you send: instructions, context, and history) and another, 5 times higher, for output tokens (what the model writes, including reasoning). Rates per million tokens, verified on October 3, 2026:

The best value option is Claude Opus 5.5. Anthropic launched it on September 22, 2026, at an official rate of $4.00 / $20.00, below the price of Opus 5 ($5.00 / $25.00), and says it performs at the level of Claude Fable 5.1 on most workloads. On Kunavo, it costs $2.80 / $14.00: less than Opus 5 ($3.50 / $17.50) or any other Opus model. Sonnet 5 ($1.40 / $7.00) costs even less. It has a 1M token context window and supports up to 128K output tokens. Its adaptive thinking is always on, and that reasoning is billed as output, so include it in your budget.

As a rule of thumb, use Haiku 4.5 for high-volume tasks (classification, extraction, summarization), Opus 5.5 or Sonnet 5 for everyday work, and Fable 5 only when you’ve tried Opus and it wasn’t enough. Rates for every model are on the pricing page.

Claude Sonnet 5: now also below the official rate

On August 31, 2026, Anthropic made Claude Sonnet 5’s launch rate, $2.00 / $10.00, its standard permanent price. Kunavo charges $1.40 / $7.00 for it.

If price per token is what matters, Sonnet 5 costs less on Kunavo than Claude Opus 5.5—$1.40 / $7.00 versus $2.80 / $14.00—and Opus 5.5 is the least expensive way to access Opus-family reasoning, priced 30% below its own official rate ($4.00 / $20.00). Claude Opus 5 costs $3.50 / $17.50, which is 30% below its official rate ($5.00 / $25.00). Claude Opus 4.8 remains available at $3.50 / $17.50.

The minimum API call

Kunavo exposes the API in both OpenAI-compatible and Anthropic Messages formats. Existing code only needs the base URL and key changed; the key starts with sk-kn-:

curl https://api.kunavo.com/v1/chat/completions \
  -H "Authorization: Bearer $KUNAVO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 300,
    "messages": [{"role": "user", "content": "Explica en tres líneas cómo se cobra la API de Claude."}]
  }'

Prompt caching: the biggest cost lever

Cached input is billed at 10 % of the input rate (at 2.5 % for Claude Fable 5.1 and at 5 % for Claude Opus 5.5): for Sonnet 5, $0.14 per million instead of $1.40; for Opus 5.5, $0.14 instead of $2.80. Writing to the cache costs 1.25 times the input rate ($1.75 per million for Sonnet 5), so it’s worthwhile only for content that gets reused: a long system prompt, documents you consult repeatedly, or conversation history. With prompts that change on every call, caching increases costs instead of reducing them. For details on how it works and why it sometimes fails through a gateway, see Prompt caching in Claude (in English) and the caching documentation.

Calculate your monthly spend

More useful than the table is a figure based on your own usage. This script calculates a month’s cost using the rates above:

# Gasto mensual de la API de Claude a partir de tus tokens reales.
# Tarifas de Kunavo en USD por millón de tokens: (entrada, salida, lectura de
# caché como fracción de la tarifa de entrada)
TARIFAS = {
    "claude-haiku-4-5": (0.70, 3.50, 0.10),
    "claude-sonnet-5":  (1.40, 7.00, 0.10),
    "claude-opus-5-5":  (2.80, 14.00, 0.05),
    "claude-opus-5":    (3.50, 17.50, 0.10),
    "claude-fable-5":   (7.00, 35.00, 0.10),
}

def gasto_mensual(modelo, entrada, salida, lectura_cache=0, escritura_cache=0):
    e, s, c = TARIFAS[modelo]
    return (
        entrada / 1e6 * e                   # entrada nueva
        + salida / 1e6 * s                  # salida (incluye el razonamiento)
        + lectura_cache / 1e6 * e * c       # lectura de caché: la fracción c
        + escritura_cache / 1e6 * e * 1.25  # escritura en caché: 125 %
    )

# Ejemplo: 30 M de tokens de entrada al mes (24 M leídos de caché) y 2 M de salida
print(gasto_mensual("claude-sonnet-5", 6_000_000, 2_000_000, lectura_cache=24_000_000))
# ≈ 25.76 USD

In this example—30 million input tokens, 24 million of them read from cache, and 2 million output tokens with Sonnet 5—the month costs about $25.76. If you don’t code, the Claude token calculator (in English) does the same calculation with a form.

3. Subscription or API? The break-even point

The answer depends not on how many days you use Claude, but on how much work you ask it to do. Two units help quantify usage: a chat exchange (about 2,000 input and 700 output tokens) and an agentic step, one round trip to the model within a coding task (about 25,000 input and 1,200 output tokens, because the context and files are resent at every step). At Kunavo’s rates, without caching:

With Sonnet 5 on Kunavo (costs $1.40 / $7.00):

• $20 per month (Pro) covers about 2,600 chat exchanges—or about 87 per day—or roughly 460 agentic steps, equal to about 23 tasks with 20 steps each.
• $100 per month (Max, entry tier) covers about 2,300 agentic steps, or 115 tasks with 20 steps each.

If your usage for a month is below those figures, the API costs less, and a month with no usage costs $0. If you exceed them month after month, the subscription is better, as long as you stay within its limits: for people who use Claude heavily every day, the plan usually costs less than paying per token.

Two caveats before you decide. These figures are pessimistic for the API because they don’t account for caching, which can substantially reduce the cost of repeated input. The app also includes something the API doesn’t: an interface. To chat through the API, you need a client such as LibreChat or Open WebUI (comparison, in English). Long conversations and those with attached documents cost more per message because the full history is resent with each one.

For teams, separate keys allocate costs by person or project. Each key can also have a monthly spending cap (after which the API returns a 402) and an allowed IP list, configured in the keys dashboard.

4. What about Claude Code?

Claude Code, Anthropic’s terminal-based coding agent, has no separate price: use it with a Pro or Max subscription (it isn’t included in the Free plan) or with an API key, which is charged per token like any other call. Its default model is Claude Opus 5.5, which costs $2.80 / $14.00 per million input / output tokens on Kunavo. See Claude Code pricing for practical costs, the coding break-even point, and setup instructions.

Paying for Claude from Spain and Latin America

The subscription is paid on claude.ai by card and renews automatically each month or year; local taxes may be added to the published price, depending on the country.

The Kunavo API balance is topped up with Stripe: cards (Visa, Mastercard, American Express, JCB, UnionPay), Apple Pay, Google Pay, and Link, among other methods depending on your country. It is a prepaid balance: minimum top-up of $10, the balance does not expire, and failed requests are not charged. Larger top-ups add extra balance: $100 gives $110 (+10%), $1,000 gives $1,200 (+20%), and $5,000 gives $6,250 (+25%). SEPA direct debit is not offered (direct debit); any debit card from your bank works normally.

For latency, taxes, and data protection by country, see the Claude API guide for Spain and Latin America; billing details are in the billing documentation (in English).

What Kunavo does not replace

Kunavo serves the Claude API over shared capacity: there is no dedicated quota or contractual SLA. If your product needs a guaranteed quota or service-level agreement, contract directly with Anthropic. And if you mainly use Claude in the app, every day and extensively, a subscription remains the cheapest option: the API makes sense when your usage is irregular, when you integrate it into your own software, or when you need to allocate and limit spending by key.

How to pay less for Claude

1. Enable prompt caching. Cached input costs 10 % of the regular rate (2.5 % for Claude Fable 5.1 and 5 % for Claude Opus 5.5); for repeated content, this is the biggest cost-saving lever on its own.
2. Choose a model for each task. An agentic step costs $0.0217 with Haiku 4.5 and $0.217 with Fable 5: do not leave the most expensive model as the default.
3. Limit the output. Output costs 5 times more than input, so max_tokens and a specific response format show up directly on the bill.
4. Do not carry history forward. Each message resends the entire conversation as input; when changing topics, start a new conversation.

Those four levers are complemented by the rate: Kunavo charges below Anthropic's official rate for the models on this page, with one key and no monthly fee. Existing code works by changing the base URL to https://api.kunavo.com/v1; for first steps, see the quickstart (in English).

Don't have a card that works with Anthropic? See how to pay for the Claude API without a credit card.

Frequently asked questions