Claude API is billed by token, and prices vary considerably by model. The table below lists Anthropic's official rates alongside Kunavo's prices (per million tokens, in USD); rates are verified as of October 1, 2026. You can top up using Alipay or WeChat Pay and pay in yuan.
Prices by model
| Model | Official input / output | Kunavo input / output | Below official rate | Cache read |
|---|---|---|---|---|
| Claude Fable 5.1 | $10.00 / $50.00 | $7.00 / $35.00 | 30% | 2.5% of input price |
| Claude Opus 5.5 | $4.00 / $20.00 | $2.80 / $14.00 | 30% | 5% of input price |
| Claude Opus 5 | $5.00 / $25.00 | $3.50 / $17.50 | 30% | 10% of input price |
| Claude Sonnet 5 | $2.00 / $10.00 | $1.40 / $7.00 | 30% | 10% of input price |
| Claude Haiku 4.5 | $1.00 / $5.00 | $0.70 / $3.50 | 30% | 10% of input price |
See the pricing page for the full list of models and prices, including GPT, image, and video models.
What a single call actually costs
Take a typical step from a coding agent: about 25,000 input tokens (code context, tool results) and about 1,200 output tokens. On Kunavo, without caching:
- Claude Fable 5.1: about $0.217 per step
- Claude Opus 5.5: about $0.0868 per step
- Claude Opus 5: about $0.108 per step
- Claude Sonnet 5: about $0.0434 per step
- Claude Haiku 4.5: about $0.0217 per step
To estimate your own usage, use the Claude token cost calculator.
How much can caching save?
When the same long prompt (system prompt, documents, codebase context) is sent repeatedly, from the second request onward, cache reads are billed at: Claude Sonnet 5 charges just 10% of the input price, while Claude Opus 5.5 charges 5%. The first write to the cache is billed at 1.25 times the input price. Most input in an agent loop is cache hits, so the actual bill is usually far below the estimate above without caching.
How to choose a model
- Claude Haiku 4.5: lowest cost (input $0.70), suitable for classification, extraction, and simple Q&A.
- Claude Sonnet 5: the default choice for everyday coding and agents, balancing speed and quality.
- Claude Opus 5.5: for harder, longer tasks and complex code.
- Fable: strongest reasoning, highest per-token price; reserve it for tasks that truly need it.
How to pay
Anthropic officially accepts only credit and debit cards. Through Kunavo, you can top up using Alipay or WeChat Pay; the minimum is $10, there is no monthly fee, and balances never expire. See the Claude API top-up guide for Alipay and WeChat Pay.
Frequently asked questions
How is Claude API pricing calculated?
It is billed by token: input (the content you send to the model) and output (the content the model generates) are priced separately per million tokens, with output costing more than input. There is no monthly fee; you pay for what you use.
How much lower are Kunavo's prices than Anthropic's official rates?
Every Claude model in the table on this page is below Anthropic's official rate, by 30%–30%, depending on the model. See the table and pricing page for details.
Can I pay in Chinese yuan?
Yes. At checkout, Stripe offers Alipay and WeChat Pay for top-ups, and amounts are shown in yuan when accessed from mainland China. See the Alipay and WeChat Pay top-up guide for steps.
Does my balance expire? Are failed requests charged?
Your balance does not expire, the minimum top-up is $10, and failed requests (4xx/5xx) are not charged.
Does Claude Code use these prices too?
Yes. When Claude Code is connected using an API key, calls are charged against your balance at the same token rates, without being subject to subscription plan usage limits.