Claude Code Pricing 2026 — Subscription or API, Real Cost, and Break-Even

Claude Code has two completely distinct pricing models, and almost all confusion about what it costs comes from mixing them up. A claude.ai subscription includes the tool in a monthly plan with usage limits. An API key bills per token, with no subscription and no cap. This page brings together current figures for both options, the calculation to help you choose, the cost of a real month of work, and how to pay from France.

Rates verified on October 3, 2026 against claude.com/pricing and Anthropic's API pricing documentation. Kunavo's per-token prices are read live from the model catalog.

The two options side by side

What the subscription costs

Claude Code is included in claude.ai's paid plans—Anthropic's setup documentation specifies that the free plan does not include access to Claude Code. Pro costs $20/month ($17/month with annual billing, $200 upfront) and covers one developer working within the plan limits. Max starts at $100/month, for those who keep hitting these limits; its tiers are presented as multiples of the Pro allowance (5x, 20x). Team and Enterprise add per-seat quotas and spending controls for administrators.

Two details matter from France: the charge is monthly and in dollars, so your bank's foreign transaction fee applies every month; and usage limits are shared with Claude chat, so an afternoon of chatting uses the same window as your coding session.

What the API route costs

There is no subscription here: you pay per token. Rates are in USD per 1M tokens, with Anthropic's public rates alongside for a direct comparison.

Cached input is billed at 10% of the input rate, and a five-minute cache write at 125%. Because Claude Code rereads the entire conversation at every step, these two multipliers have the biggest effect on your actual bill—more than your choice of model.

The real cost of one step

An “agentic step” is one back-and-forth with the model within a task: system prompt + conversation + file context in, an edit or explanation out. Measurements put this at about 25,000 input tokens / 1,200 output tokens. At Kunavo rates:

The break-even point

It depends on the number of steps, not the number of days worked. At $0.043 per step on Sonnet 5, a $20 Pro subscription is equivalent to about 460 steps per month, or around 23 tasks of 20 steps each. Below that, pay-as-you-go is cheaper, and months without coding cost nothing; above it, the subscription is better until you hit the plan limits.

To use your own figures, run /usage in Claude Code and enter the four numbers:

# Ce qu'a coûté une session Claude Code, à partir de ses tokens.
# Lancez /usage dans Claude Code pour récupérer les quatre nombres réels.
RATES = {  # USD par 1M de tokens chez Kunavo : (input, output)
    "claude-haiku-4-5":  (0.70, 3.50),
    "claude-sonnet-4-6": (2.10, 10.50),
    "claude-sonnet-5":   (1.40, 7.00),
    "claude-opus-5":     (3.50, 17.50),
    "claude-opus-5-5":   (2.80, 14.00),
}

def cout_session(model, inp, out, cache_read=0, cache_write=0):
    i, o = RATES[model]
    return (
        inp / 1e6 * i                   # input neuf
        + out / 1e6 * o                 # output, tokens de raisonnement inclus
        + cache_read / 1e6 * i * 0.10   # lecture de cache : 10% du tarif input
        + cache_write / 1e6 * i * 1.25  # écriture de cache 5 min : 125%
    )

print(cout_session("claude-sonnet-5", 1_200, 5_300, 940_000, 50_000))

Point Claude Code to Kunavo

Claude Code works with any Anthropic-compatible endpoint. Two environment variables are all you need to switch to pay-as-you-go without changing your installation or workflow:

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

Keep all four model lines. Claude Code’s default model and its opus alias point to the latest Opus, and if Kunavo does not serve it yet, the first request returns a 404 error. The ANTHROPIC_DEFAULT_OPUS_MODEL line pins the alias to Opus 5.5 (claude-opus-5-5), which requires Claude Code v2.1.280 or later (otherwise, run claude update). The sonnet alias requests Sonnet 5.5, which Kunavo does not serve: without ANTHROPIC_DEFAULT_SONNET_MODEL, /model sonnet, the execution phase of opusplan, and subagents configured for sonnet fail with a 404 error. The key is generated from /app/keys. The complete step-by-step instructions are in the key configuration guide, and if you use Kilo Code instead of the CLI, see configure Kilo Code with Claude.

Four ways to pay less

1. Let prompt caching work. Cached input costs 10% of the input rate, and Claude Code rereads the whole conversation at every step, making this the single most effective lever. See the cache documentation.
2. Clear context between unrelated tasks. Carrying one long session forward makes input tokens grow at every step.
3. Choose the model based on the size of the problem. Do not default to the biggest model: the table above shows an order-of-magnitude difference per step between Haiku and Opus.
4. Route internal calls to Haiku. With ANTHROPIC_DEFAULT_HAIKU_MODEL, Claude Code's auxiliary calls to itself can use the cheapest model.

Paying from France

The subscription is a monthly dollar charge on a card. The API-key route uses a Stripe checkout that accepts cards (Visa, Mastercard, Amex), Apple Pay, Google Pay, and Link for one-click top-ups. Pay by Bank is not available from France: in this checkout, it is offered only to buyers in the United Kingdom, Ireland, and Finland (at Stripe, France is still in the private-preview stage). This is a prepaid wallet: the minimum top-up is $10, and the balance does not expire—a month without coding triggers no charge. Details are in the billing documentation.

Comparing this with OpenAI's agent? The same calculation for Codex is in Codex pricing.

Frequently asked questions