Models

Use an open model from the catalog or deploy your own. Both use the same inference API.

Available models

Rates are in USD per million tokens. Select a model for its context limit, serving configuration, and input, output, and cache rates. See pricing for billing details.

Each model page lists reasoning, tool calling, structured output, streaming, and supported input and output types. “Not verified” means the revision has no recorded capability check.

Using a model

Set model in your inference request to a public model ID or your private deployment’s name.

Private names resolve within the organization your API key belongs to. A repository ID selects the public model, not your private deployment of that repository.

GET /v1/catalog/models lists the public catalog without a key. Use data[].id in inference requests. Your deployments are listed separately by GET /v1/models, which requires models:read and includes pending and disabled models. Use their returned slug for inference.

Pinning a revision

Use support-agent@<revision_id> to select a checked Boardwalk serving revision, including a private candidate that was checked but not yet promoted. An unavailable model or revision returns 404; Boardwalk never substitutes a different model for an unavailable pin.

A revision pin fixes the source weights, runtime, and serving settings. The request receipt records the exact serving configuration used. Get revision IDs from GET /v1/models/{id}/revisions, or use config_revisionfrom the public catalog. Stopped models cannot serve pinned requests.