Cheap Claude API — Opus, Sonnet and Haiku from $0.13/1M
Every Claude model through one OpenAI-compatible key — up to 87% below Anthropic's official price. The Anthropic-native endpoint works too, so extended thinking, native tool_use and prompt caching arrive intact — Claude Code plugs in by changing a single environment variable. Pay in ₽, USDT or by card, no VPN needed.
Claude API pricing
USD per 1M tokens, next to Anthropic's official rate. Updated automatically from our live price list.
| Model | Input | Output | Anthropic | Savings |
|---|---|---|---|---|
| Claude Fable 5 | $3.80 | $19 | $10 | −62% |
| Claude Fable 5.1 | $3.80 | $19 | $10 | −62% |
| Claude Opus 5 | $0.66 | $3.33 | $5.00 | −87% |
| Claude Opus 4.6 | $0.66 | $3.30 | $5.00 | −87% |
| Claude Opus 4.7 | $0.66 | $3.30 | $5.00 | −87% |
| Claude Opus 4.8 | $0.66 | $3.30 | $5.00 | −87% |
| Claude Opus 5.5 | $0.53 | $2.65 | $4.00 | −87% |
| Claude Sonnet 4.6 | $0.40 | $2.00 | $3.00 | −87% |
| Claude Sonnet 5 | $0.26 | $1.33 | $2.00 | −87% |
| Claude Sonnet 5.5 | $0.26 | $1.33 | $2.00 | −87% |
| Claude Haiku 4.5 | $0.13 | $0.65 | $1.00 | −87% |
Why take Claude from here
Anthropic-native, not a wrapper
Requests go to /v1/messages as passthrough, so extended thinking blocks keep their signatures and native tool_use survives multi-turn agent runs. Gateways that convert Claude to the OpenAI format break both.
Prompt caching that actually bills as cache
cache_control on 5-minute and 1-hour windows works, and cache reads are billed at the cache rate — the price list below shows both the read and the write column per model.
No model swapping
response.model always equals what you asked for. We don't quietly route a request for the flagship to a cheaper tier — check the field yourself on any call.
How to connect Claude
Claude Code
Base URL goes WITHOUT /v1 — the SDK appends /v1/messages itself. With /v1 you get /v1/v1/messages and a 404.
export ANTHROPIC_BASE_URL="https://cheapai.io"
export ANTHROPIC_AUTH_TOKEN="sk-..."
claudeAny OpenAI SDK
Works in Cursor, Cline, Kilo Code, LangChain and anything else that lets you set a base URL.
from openai import OpenAI
client = OpenAI(
api_key="sk-...", # from cheapai.io/keys
base_url="https://cheapai.io/v1",
)Also available without a CDN: https://api.cheapai.io/v1 — use it for long requests where the model stays silent over ~2 minutes, or where the CDN is unreachable.
Full guides for every tool: documentation →
FAQ
How much does the Claude API cost here?
From $0.13 per 1M input tokens. 11 Claude models are live right now and the table above is pulled from the same price list the billing uses, so it is never stale.
Is this the real Claude, or a cheaper model behind the name?
The real thing. response.model always matches the model you requested, so you can verify it on any call without taking our word for it.
Do extended thinking and prompt caching work?
Yes — requests reach Anthropic's native /v1/messages as passthrough, so thinking blocks keep their signatures across tool calls and cache_control is honoured and billed at cache rates.
Do I need a VPN?
No. The domain and the API are reachable worldwide, and you can top up in ₽, USDT or by card.
Is there a subscription?
Not required. The default is pay-as-you-go from a balance that doesn't expire; monthly plans exist separately and give up to 40% off, but you never have to take one.