Choose a model
Start with the credential-scoped catalog
Section titled “Start with the credential-scoped catalog”Use the public model ID shown for your organization in the Developer Portal or returned by GET /v1/models. Availability is specific to the calling credential, so discover the list with the credential you intend to use.
When supported on your platform, retrieve one accessible model with GET /v1/models/{model_id}. The response describes the model from the caller’s perspective; it does not expose provider credentials or make an inaccessible model usable.
Gateway accepts published, provider-neutral model IDs. Provider-native identifiers, provider prefixes, aliases, slash-qualified names, and internal routes are outside the user-facing API. Get IDs from the Developer Portal or the Gateway model endpoint.
Check before you deploy
Section titled “Check before you deploy”The Developer Portal model details show what you need to choose a client:
- access status and lifecycle status;
- supported public transports;
- capabilities, context window, and maximum output;
- recommended coding harnesses and use cases;
- per-harness compatibility status and feature tier;
- organization notes and model advisories; and
- customer-facing rate information when available.
Use Check model compatibility to interpret Supported, Limited, and Unavailable harness statuses. A model can appear in a broad catalog but remain unavailable to your credential, transport, route, capability, or pricing policy.
Let the Gateway select a model
Section titled “Let the Gateway select a model”When a supported request offers it, model: "auto" asks Gateway to select one eligible public model. Selection follows your organization’s access, the request’s supported capabilities, route readiness, and pricing requirements. It does not bypass policy, allowances, or capability checks.
Use an explicit public model ID when a workload requires a specific model, predictable behavior, or a transport-specific feature.
Model availability is not provider access
Section titled “Model availability is not provider access”Organization administrators curate AI Models and choose preferred and fallback provider routes in the organization’s model configuration. Callers do not configure provider-native targets or provider credentials. A preferred route can be unavailable even when the public model remains visible; inspect the model details and returned request diagnostics before changing client configuration.