Claude API Pricing 2026 — Model Rates, Payment, and Real Costs

Claude is the standard for reasoning, code, and agents, and Kunavo serves every Claude model 30% below Anthropic’s official prices through a single OpenAI-compatible API. This also applies to claude-sonnet-5: Anthropic made the $2.00 / $10.00 rate permanent on August 31, 2026, and Kunavo is 30% below. This guide brings together what French developers search for most—the Claude API price, payment, and real-world costs: current rates by model, how to pay from France, VAT, cost examples you can verify yourself, and the biggest ways to lower your bill.

Claude API prices at a glance

Rates are in USD per million tokens, as charged by Kunavo. The “Anthropic rate” column is Anthropic’s published price for the same model, for a direct comparison.

Live rates are always available on the pricing page and on each model page (claude-sonnet-4-6, claude-haiku-4-5, claude-opus-4-7, claude-opus-5-5). Haiku is the budget option, Sonnet 4.6 is the previous generation, Sonnet 5 is the near-Opus-tier option for code (and, on Kunavo, the least expensive Sonnet), Opus 5.5—the newest Opus and the least expensive on Kunavo—is for difficult reasoning, and Fable 5 is Anthropic’s most capable model.

Claude Opus 5.5 is now the best value among Opus models. Released on September 22, 2026, Anthropic lists it at $4.00 / $20.00 per 1M tokens, below Opus 5 ($5.00 / $25.00), and says it performs at the level of Claude Fable 5.1 on most tasks. On Kunavo, it costs $2.80 / $14.00, ~30% below Anthropic’s rate: cheaper than Opus 5 ($3.50 / $17.50) and any other Opus. Sonnet 5 ($1.40 / $7.00) now costs even less. A long-context analysis with 50,000 input tokens and 2,000 output tokens costs $0.21 on Opus 5 and $0.17 on Opus 5.5. Adaptive thinking is always active on Opus 5.5: allow for its thinking tokens in the output line.

Claude Opus 5 remains available. Anthropic lists it at $5.00 / $25.00 per 1M tokens—half the rate of Fable 5 ($10.00 / $50.00) for near-flagship reasoning. On Kunavo, it costs $3.50 / $17.50, ~30% below Anthropic’s rate and exactly the rate of the older claude-opus-4-7: moving a workload from Opus 4.7 to Opus 5 means changing one word in the model name, with no extra cost.

How to pay for the Claude API from France

Payment is processed through Stripe. There is no subscription or Anthropic Console setup: you top up a balance, and each call is charged at the per-token rates above. Payment methods available in France:

• Cards: CB, Visa, Mastercard, American Express
• Apple Pay, Google Pay, and Link (one-click payment)

SEPA direct debit does not appear at checkout. A debit card linked to your SEPA account works normally. Pay by Bank is not offered in France, where it exists only in private preview; Stripe reserves it for buyers in the United Kingdom, Ireland, and Finland.

Prices are in USD; Stripe may convert the amount to EUR at checkout, and if you pay in USD, foreign exchange fees depend on your bank. Minimum top-up: $10 USD, pay-as-you-go, and since this is a prepaid balance, the balance never expires. Regarding VAT: the standard checkout does not show VAT or ask for a VAT number; for a receipt, email contact@kunavo.com.

Cost examples

Real figures calculated using Kunavo rates:

Processing 10,000 support tickets per day on Haiku costs about $12.60/day. Here is the calculation so you can run it yourself:

# Tarifs Claude sur Kunavo (USD par 1M de tokens) : (input, output)
RATES = {
    "claude-haiku-4-5":  (0.70, 3.50),
    "claude-sonnet-4-6": (2.10, 10.50),
    "claude-opus-4-7":   (3.50, 17.50),
    "claude-opus-5":     (3.50, 17.50),
    "claude-opus-5-5":   (2.80, 14.00),
}

def cout(model: str, in_tokens: int, out_tokens: int) -> float:
    i, o = RATES[model]
    return in_tokens / 1_000_000 * i + out_tokens / 1_000_000 * o

print(cout("claude-haiku-4-5", 800, 200))      # un ticket support     -> $0.00126
print(cout("claude-sonnet-4-6", 6_000, 500))   # une réponse RAG       -> $0.0179
print(cout("claude-opus-4-7", 50_000, 2_000))  # analyse long contexte -> $0.21
print(cout("claude-opus-5", 50_000, 2_000))    # analyse Opus 5        -> $0.21
print(cout("claude-opus-5-5", 50_000, 2_000))  # analyse Opus 5.5      -> $0.168

French uses about 20 to 40% more tokens than English for the same content (accents, longer words, finer tokenizer segmentation)—factor that into your budget.

Prompt caching—the biggest cost saving

If your system prompt or retrieved context is large and stable, prompt caching charges cached input at 10 % of the input rate (2.5 % for Claude Fable 5.1 and 5 % for Claude Opus 5.5). For a long, repeated system prompt, this cuts input costs by 60 to 90%. Just add one field to the request—see the in-depth Anthropic prompt caching guide and the Messages API reference.

One key for Claude, GPT, images, and video

A single Kunavo key (sk-kn-…) calls Claude as well as GPT, image generation, video, and audio—all in OpenAI-compatible format and charged to a single Stripe balance, with no provider invoices to reconcile. Claude is served through both the OpenAI-compatible endpoint /v1/chat/completions and the native Anthropic Messages API (/v1/messages), so you can keep your current SDK. The steps are in the authentication documentation.

Which Claude model should you choose?

The difficulty-based routing pattern (simple tasks go to a cheaper model) is covered in the AI cost optimization guide. For a comparison with OpenRouter, see Kunavo vs. OpenRouter.

Use Claude in VS Code—Kilo Code, Cline, Roo Code

If you use Claude through a coding agent such as Kilo Code, Cline, or Roo Code, the setup is the same: provider “OpenAI Compatible,” base URL https://api.kunavo.com/v1, your key, and a Claude model slug. The full step-by-step instructions (including the real cost of an agent session) are in the guide Set up Kilo Code with the Claude API.

Reduce Claude API costs

1. Start with prompt caching. Cached input is billed at 10 % of the input rate (2.5 % for Claude Fable 5.1 and 5 % for Claude Opus 5.5) — the best return on effort.
2. Move down a tier. Send easy tasks to Haiku; save Opus for the difficult minority.
3. Cap output. Set max_tokens — output costs about 5 times more than input.
4. Trim input. Retrieve fewer, more relevant RAG chunks instead of the entire knowledge base.

Frequently asked questions