Claude Pricing [2026 Edition] — Monthly Plans, API Rates, Claude Code

Claude pricing falls into 3 categories. The chat subscription on claude.ai is Free at $0, Claude Pro at $20 per month ($17 per month equivalent with annual billing), and Claude Max starting at $100 per month. The API has no monthly fee and is billed by token; on Kunavo, Claude Sonnet 4.6 costs $2.10 input / $10.50 output per 1M tokens — compared with Anthropic's list price of $3.00 / $15.00, that's 30% lower, depending on the model. Claude Code has no separate price; it uses either a subscription or an API key.

Most confusion over “how much does it cost?” comes from mixing up these three categories. This page separates them, shows the current figures, and calculates which is cheaper for the same usage: a subscription or the API. Many pages list plans, but few put numbers to the point where they intersect.

Rates verified as of October 10, 2026: subscription prices are from claude.com/pricing, and API list prices are from Anthropic's API pricing page. Kunavo rates are rendered from this site's catalog, so they match actual charges.

Claude pricing in 3 tiers — at a glance

1. Claude monthly plans — Free / Pro / Max / Team

A claude.ai subscription combines chat, Claude Code, and other features into a fixed fee with usage limits. Limits are managed using a rolling 5-hour window and weekly caps, and chat and Claude Code share the same limits.

Claude Pro pricing — $20/month, when to upgrade to Max

Pro costs $20/month, or $17/month billed annually ($200 upfront). What matters more than the $3 monthly difference is that annual payments are nonrefundable if you cancel early. Start with monthly billing for 1–2 months and see how often you hit the 5-hour window limit before switching to annual billing. That gives you more information to make your decision.

The signal to upgrade to Max (from $100/month) is frequency.If you hit the limit several times a week and work comes to a halt, the $80 difference pays for itself in time saved. If you only hit it about once a month, it's cheaper to switch to an API key only when you hit the limit. Your options after hitting the limit are covered in Claude Pro limits.

Is Claude free to use? Free plan limits

You can use chat for free on the Free plan.However, there is a cap on the number of messages you can send in each 5-hour window, and that cap is lower during busy periods. Claude Code is not included in the Free plan.

There is no permanent free tier for the API.You pay for what you call, but there is no monthly fee, so your bill is $0 in months when you don't use it. The practical answer to “I want to try it for free” is either chat on the Free plan or add a balance with the $10 minimum top-up and make only the calls you need.

2. Claude API pricing — price per 1M tokens by model

The API is billed on a per-token basis. Input and output tokens have separate rates, and output costs more. The rates below are per 1M tokens as of October 10, 2026.

Claude Opus 5.5 pricing — the cheapest Opus on Kunavo

Claude Opus 5.5 (claude-opus-5-5) is an Opus model released by Anthropic on September 22, 2026. Anthropic's list price is $4.00 / $20.00 per 1M tokens, lower than the $5.00 / $25.00 for Opus 5. Anthropic says it “performs at the same level as Claude Fable 5.1 for most tasks.” On Kunavo, it costs $2.80 / $14.00 (about 30% below list price). It costs more than Sonnet 5 ($1.40 / $7.00), but less than Opus 5 ($3.50 / $17.50), making it the lowest-priced Opus model available on Kunavo.

There are two things to keep in mind when estimating costs. First, cached reads cost 5% of the input rate (for most models, 10%), making repeated content more cost-effective. Second, thinking: Opus 5.5 always has adaptive thinking enabled, so include thinking tokens when estimating output. It has a 1,000,000-token context window and a maximum output of 128,000 tokens, and is also the default model in Claude Code.

Claude Sonnet 5 pricing — below list price since September 24

Since September 24, 2026, Claude Sonnet 5 has also been available below list price on Kunavo.After Anthropic made its introductory discounted rates of $2.00 / $10.00 the permanent standard price, Kunavo charged the same rates. Now it costs $1.40 / $7.00 (about 30% below list price). If you need Opus-level reasoning, the cheapest Opus on Kunavo is Claude Opus 5.5 ($2.80 / $14.00). If Sonnet 5 is enough for the task, it costs even less.

Example calculation you can verify yourself

Numbers calculated from your own usage are more useful than a price table. The code below calculates a full month of usage.

# 1 か月の Claude API 料金を、実際のトークン数から出す。
# レートは 1M トークンあたり USD (input, output)。3 つめはキャッシュ読み出しの
# 比率（入力レートに掛ける）で、モデルによって違う。
RATES = {
    "claude-haiku-4-5":  (0.70, 3.50, 0.10),
    "claude-sonnet-4-6": (2.10, 10.50, 0.10),
    "claude-opus-5":     (3.50, 17.50, 0.10),
    "claude-opus-5-5":   (2.80, 14.00, 0.05),
}

def monthly_cost(model, inp, out, cache_read=0, cache_write=0):
    i, o, r = RATES[model]
    return (
        inp / 1e6 * i                   # 新規入力
        + out / 1e6 * o                 # 出力（thinking トークンを含む）
        + cache_read / 1e6 * i * r      # キャッシュ読み出し: 入力レート × r
        + cache_write / 1e6 * i * 1.25  # 5 分キャッシュ書き込み: 125%
    )

# 例: 月 3,000 万入力トークン / 200 万出力トークン、うち 8 割がキャッシュ読み出し
print(monthly_cost("claude-sonnet-4-6", 6_000_000, 2_000_000, cache_read=24_000_000))

Here are the actual costs for some representative units (Claude Sonnet 4.6, Kunavo rates).

The smallest call looks like this.

curl https://api.kunavo.com/v1/messages \
  -H "x-api-key: $KUNAVO_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 512,
    "messages": [{"role": "user", "content": "料金体系を 3 行で説明して"}]
  }'

Prompt caching — the most effective lever

Cached input is billed at 10% of the input rate（5% on Claude Opus 5.5, 2.5% on Claude Fable 5.1）.When you repeatedly send the same content, such as a long system prompt, documents reread on every call, or conversation history, this alone delivers the biggest savings. Writes cost 125% of the input rate with a 5-minute cache, so they only pay off when the cached content is read on the second or a subsequent use. Adding caching to prompts that change every time can therefore cost more.

How it works and the typical reasons caching gets lost through a gateway are covered in Claude prompt caching; billing details are in the caching documentation.

3. Claude Code pricing — subscription or API key

Claude Code itself is free.The CLI, editor extension, and settings are all free; you pay only for model calls. There are two ways to pay: a subscription (included in the monthly Pro / Max price, with usage limits) or an API key (usage-based billing with no monthly fee or usage limits).

Subscription vs API key break-even point

The break-even point depends on the number of runs, not the number of workdays. One agentic step (25,000 input / 1,200 output tokens) costs about $0.0651 with Claude Sonnet 4.6. So:

• $20/month (Pro) ≈ 310 steps ≈ about 15 tasks with 20 steps each
• $100/month (entry-level Max) ≈ 1,500 steps
• For chat-focused use, $20/month ≈ about 1,700 exchanges

These estimates use Claude Sonnet 4.6. Claude Opus 5.5 is the default model in Claude Code; running it directly with a Kunavo API key costs about $0.0868 per step, or about 230 steps for $20/month.

Below this level, usage-based billing is cheaper, and months with no activity cost $0. If you consistently exceed it, a subscription is more cost-effective until you hit the plan's usage limit. Teams can separate costs by person and purpose by using separate keys. Usage-based billing is also easier to account for in months when seats are unfilled or for usage not tied to an individual, such as CI.

Detailed Claude Code calculations, environment variable setup, and how to continue after hitting a usage limit are covered in Claude Code pricing and How to use Claude Code. A side-by-side comparison using the same calculation method as Codex is in Codex pricing.

How costs change when you move to a lower tier

If your use case is mostly “large-scale classification and summarization,” switching models will save more than any other cost-saving measure. As of October 10, 2026, Claude Haiku 4.5 costs $0.70 input / $3.50 output per 1M tokens, while Claude Sonnet 4.6 costs $2.10 / $10.50. Routing steps that don't need high-quality reasoning to a lower tier can greatly reduce your bill. See the Chat Completions documentation for how to implement model switching.

Paying for Claude from Japan

Subscriptions are recurring charges, and on the web version you can pay only by credit or debit card (through the app, App Store / Google Play payment methods are available). The Kunavo API balance uses Stripe Checkout and supports cards including JCB (Visa, Mastercard, American Express, JCB, UnionPay), Apple Pay, and Google Pay. It is a prepaid wallet system with a minimum top-up of $10, and the balance does not expire. There is no charge for months in which you make no calls, and failed requests are not charged. Convenience-store payments and PayPay are not supported. A comparison of payment methods for subscriptions, Console, and Kunavo is compiled in Claude payment methods.

Required disclosures for domestic businesses are in Disclosures under the Act on Specified Commercial Transactions; billing details are in the billing documentation. Steps to make your first API call are in Quickstart.

Lower your Claude costs — in order of impact

1. Use prompt caching.Cached reads cost 10%（5% on Claude Opus 5.5, 2.5% on Claude Fable 5.1） of the input rate. If your setup repeatedly sends the same content, this is the single biggest lever.
2. Choose a model to fit the task.Don't default to the top-tier model. Haiku is often enough for classification, extraction, and formatting, and the difference between $0.0217 and $0.11 per step adds up directly.
3. Limit output.Output tokens cost more than input tokens, so max_tokens and specifying an output format have a direct effect.
4. Don't accumulate history unrelated to the input.Clear the conversation when you move to a different task. Long histories are billed as input in full on every call.

On top of that, using an endpoint priced below Anthropic's list rates stacks with all four measures above. Kunavo provides both OpenAI-compatible and Anthropic-compatible interfaces, so existing code works by changing only base_url. See Quickstart for setup and Model catalog for rates for all models.

Pricing for image and video models is covered on separate pages: Nano Banana Pro pricing and Veo 3 pricing.

Frequently asked questions