One API.Every top AI model.
Reach GPT, Claude and Grok through a single OpenAI-compatible endpoint.
Test it with $0.20 on us.
Sign up and the credit is already there, so you can run real requests against real models before you pay anything.
The things you'd want to check before paying
We don't store your prompts
Our usage records hold token counts, model name, timing and cost — nothing from the body of your request or the model's reply. The content is never written to our database.
Credits don't expire
Top up once and the balance stays yours. Unused balance is refundable within 7 days, sent back in USDT.
A person answers
Support is a Telegram chat with someone who can actually look at your account — not a ticket queue. Usually a few hours.
The models you actually use
One key. Switch models by changing a single string.
Image generation, billed per picture
GPT Image 2.5 (plus Flare and Sunburst) and GPT Image 2 through the same OpenAI-compatible endpoint and the same balance. Flat per image — no token arithmetic, no subscription. Image models sit in their own key group, so they need a key of their own.
1K — $0.0536
About 1.6 megapixels per image, usually ready in around 25 seconds.
Five shapes, one price
1:1 1254×1254 · 3:2 1536×1024 · 2:3 1024×1536 · 16:9 1672×941 · 9:16 941×1672
Failed requests are free
Requests that error out or are stopped by the content filter are not charged.
Calling the API yourself? Keep the long edge of the size you send at 1024 — for example 1024x576 for 16:9 or 576x1024 for 9:16. It is billed at the 1K price and still comes back at full size. Full price list
Built for people who ship
Drop-in compatible
Works with the official OpenAI and Anthropic SDKs. Change your base URL and keep the rest of your code.
Made for coding agents
Verified with Codex, Claude Code and Cherry Studio. Streaming, tool calls and long context all work.
Pay as you go
No subscription. Top up, spend what you use, and see every request in the console.
What we store, and what never reaches our database
No badge wall and no borrowed logos. Here is the actual record a request leaves behind.
Your prompts are not stored
Our database has no column for request or response content. A call writes a timestamp, the model name, token counts, the cost, the key that made it and the source address. That row is the entire record, and it is what your invoice is built from.
Every charge is itemised
Input, output, cached input and cache writes are logged separately for each call, with the exact amount deducted and the rate applied. Any line on your bill can be traced back to the request that produced it.
Nothing holds you here
A standard OpenAI-compatible endpoint, prepaid credit that never expires, and unused balance refundable in USDT within seven days of a top-up. Point your base URL back at the provider and the same code keeps running.
Three steps to your first call
Create an API key in the console
Point your SDK at our endpoint
Call any model with the same code
// OpenAI SDK — only the base URL changes import OpenAI from "openai"; const client = new OpenAI({ apiKey: process.env.APICLAN_KEY, baseURL: "https://apiclan.us/v1", }); const r = await client.chat.completions.create({ model: "gpt-5.6", messages: [{ role: "user", content: "Hello" }], });
Using a desktop app instead?
Not calling it from code? Claude Desktop and the Codex app have no base-URL box of their own, so the configuration is written by CC Switch: what to enter for each one