Models API
Use the Models API to discover models before choosing an inference endpoint. The list operation is OpenAI-compatible and returns compact identity records. Retrieve one model for SandBase capability, schema, and pricing metadata.
| Operation | Endpoint | Documentation |
|---|---|---|
| List models | GET /v1/models | List Models |
| Get one model | GET /v1/models/{id_or_name} | Get Model |
| Invoke an API capability by path | GET/POST /v1/api/{vendor}/{upstream_path} | API Passthrough |
| Run a model or API capability | POST /v1/run | Run Capability |
| Get an asynchronous result | GET /v1/run/{id} | Get Run Result |
| Receive an asynchronous callback | webhook_url on POST /v1/run | Webhook callbacks |
| Browse supported models | — | Supported Models |
| Compare capabilities | — | Capabilities |
To run a model, use the Model API Reference for normalized Chat Completions, Anthropic Messages, image, video, and audio contracts. This page documents discovery, the unified capability runner, API passthrough, assets, and task-cost lookup.
GET /v1/models supports q, vendor, type, and order; it is not paginated. Omitting type defaults to llm, while an explicitly empty type= lists all public model types. The vendor filter is exact and case-sensitive; q is a case-insensitive substring search. Each list item contains id, object, created, and owned_by, where id is the logical model name used in requests.
Model names may contain /. Pass those slashes as path separators when calling the detail endpoint, for example /v1/models/openai/gpt-5.6-luna; the server route consumes the remaining path as one logical identifier. Discover the current ID from GET /v1/models instead of copying an identifier from an older guide.
For an asynchronous 202 response from an inference API, poll GET /v1/run/{id} with the returned opaque ID. Do not infer an ID prefix or construct a different polling path from compatibility headers.