Developer FAQ

Short answers to common API questions. If yours isn't here — check the API reference or error codes.

Is this really a drop-in replacement for the OpenAI API?

Yes. Change base_url to https://api.cheapai.io/v1 and the key to cai-... — leave the rest of the code alone. Endpoints, parameters and response format are the same. Provider-specific parameters are passed to the model as-is; if a specific model doesn't support something — it returns the same error it would directly.

Which models are available?

The catalog spans many models — OpenAI, Anthropic, Google, DeepSeek and more. The popular ones with prices are listed on the catalog page; the model name from the catalog goes into the model field. The list and prices can change — check the catalog for the source of truth.

Does streaming work?

Yes — "stream": true, response comes in chunks as SSE, stream ends with data: [DONE]. Example in the reference.

Are function calling, tool use and vision supported?

That depends on the model, not on the gateway: a request with tools / tool_choice or with images in content is forwarded to the provider, and the model responds in its own format. Model capabilities are visible from the tags in the catalog (chat, vision, embeddings, etc.).

Does one key work with all models?

Yes. The key is not bound to a model or provider — one key can call gpt-4o or claude-3-5-sonnet equally. Balance is shared across the account; you can have multiple keys (convenient per project) and revoke them individually — see API keys.

Do I need a VPN?

No. Requests go to api.cheapai.io — models respond through the gateway. This works for the API and for CLI agents like Claude Code and Codex.

What rate limits apply?

Each model has its own requests-per-minute limit. On exceeding it you get 429 — retry with exponential backoff (official SDKs do this automatically). Details — Error codes.

How is request cost calculated?

From the tokens in the usage field of the response and the model tariff (input and output separately). No subscription — you only pay for what you use. Tariffs and formula — on the Pricing page; the cost per request is in your usage history.

A request returned an error — where do I look?

First — the HTTP status and the field in the response body: that's where the specific reason lives. Decoding the statuses and what to do with each — on the page.