The documentation is live
Nine pages covering the shortest path to a response, the words the product uses, and what the API supports today.
- Published
Update, 1 October 2026: Pricing now publishes v1 rate references. Billing covers model-specific cache pricing, the Batch multiplier and capacity commitments. The article below describes the August launch.
The documentation is at /docs. It is organised in three groups, and the order is the order we think you should read them in.
Start here#
Get started is the shortest path to a response: generate a key, change one line, run the call you already have. Concepts defines the words the rest of the pages use: run mode, lane, pinning, cold start. Those four terms carry most of the meaning and guessing at them makes every later page harder. OpenAI compatibility tells you what is supported today, so you can check a parameter before you depend on it.
Some pages deliberately say less#
Several pages carry a visible marker saying which part of them is incomplete. That is not an
oversight. Models lists the catalog but not what a request costs, because there
was no published rate card at launch. Smart Router notes that grid-auto is not
currently returned by GET /v1/models, so you cannot discover it from the catalog listing.
A page that reads complete and is not costs you more than a page that tells you where the edge is.