The AI Gateway is a unified, OpenAI-compatible API for models from OpenAI, Anthropic, Google, xAI and DeepSeek — one endpoint, one token, one balance. Open the AI Gateway page in my.cubepath.com.
The value is in what it removes: no separate account, key, billing relationship and SDK per provider. You keep one integration and change providers by changing a model name.
Get started
- 1Create an API tokenClick Create API Token to get a key for the gateway.
- 2Add AI FundsTop up your AI Funds balance. Usage is pay-as-you-go, billed per token.
- 3Point your SDK at the gatewayUse any OpenAI-compatible SDK: set the Base URL to the gateway and use your CubePath API token, then pick a model by name and send requests.
Because the API is OpenAI-compatible, most existing code needs only the base URL and the key changed — the request and response shapes stay the same.
Models
Browse the Models catalog for what's available and what it costs. Each entry shows:
- Price per 1M tokens, input and output separately. Output is usually several times more expensive than input, which is why verbose responses cost more than long prompts.
- Context window — how much you can send in one request.
- Max output — the ceiling on a single response.
- Capabilities — streaming, vision, tool use.
TipSwitch providers by changing only the model name in your request. Your token, endpoint and billing stay exactly the same, which makes it cheap to A/B a cheaper model against your current one on real traffic.
Watching the cost
The dashboard tracks requests, tokens and cost, daily and per model, over the last 30 days.
Read it per model rather than in total. The usual finding is that one endpoint in your application — a summarizer, a classifier running on every request — accounts for most of the bill, and moving just that one call to a smaller model changes the number materially.
NoteBilling is per token, not per request. A single request with a large context costs far more than several small ones, so trimming what you send is often the biggest lever available.
Good practice
- Set the smallest model that passes your evals, not the biggest one you can afford. The gap in quality is usually smaller than the gap in price.
- Cap max output on endpoints where a runaway response would be expensive.
- Keep prompts out of the loop where you can. Caching a result you'll need again is free; regenerating it isn't.