SetCloud
✦ Ask AI Get started

Models

The Gateway puts the available model catalog behind one API. This page covers listing what’s available, picking a model for a task, and where your own fine-tuned models show up.

Listing models

GET /v1/models returns every model your key can call, including any custom models you’ve fine-tuned:

curl https://setcloud.com/v1/models \
  -H "Authorization: Bearer $SETCLOUD_API_KEY"

If the key carries a model allow-list, the response is filtered to it — you only ever see what that key is permitted to call.

Catalog updates

SetCloud refreshes model availability and standard token pricing every six hours, including once when the control service starts. Newly available active chat models enter the catalog after their supported API family, invocation path, account availability, and published pricing are checked. Models that disappear from discovery are disabled after two complete refreshes; the last successful catalog remains available if an upstream refresh fails.

Models rejected during live account verification stay unavailable until a successful recheck. Models that need additional account access, have a future availability date, or have unverified pricing remain unavailable. Compatibility updates trigger a fresh catalog check when the service starts. The dashboard and public rate card use the same model rates. API keys and custom models belong to the selected organization.

Token charges use reported input and output usage, apply the platform margin, and round up to whole cents per request (minimum $0.01 for nonzero usage). Text-length estimates are never used to charge an interrupted stream. A temporary credit-ledger failure retains the measured charge for an idempotent retry.

Choosing a model

The catalog spans a range of sizes and capabilities, from small, fast models for simple or high-volume calls to larger models for harder reasoning tasks. There’s no wrong model to start with — swap the model field in a request and nothing else about your integration changes, so it’s cheap to try a smaller or larger option and compare.

Models for agents and tool calling

The tools capability identifies models that can request actions. Agents uses the same catalog and rates as AI Gateway. Use Refresh models in the agent builder after a catalog update. Your saved model choice stays in place until you change it.

With the AI SDK, use chat() or chatStream() for tool calls. Complete replies include tool_calls; streaming chunks preserve incremental tool arguments. Assemble arguments by their tool-call index before parsing the JSON. The SDK returns requests for actions; your application decides whether to execute them.

Use max_completion_tokens to set the completion limit. Sampling controls may be omitted for models that do not support them.

Long-context pricing

Some models apply a higher rate to the entire request once its input crosses a threshold. The model card in AI Gateway, the agent model picker, and the public price table show this condition. Gateway and Agents use the same calculation.

Use the condition shown for the selected model in the live catalog. A model name in an older example is not evidence that your key can call it.

Your fine-tuned models

Once a fine-tuning job finishes and its model deploys, it appears in GET /v1/models as a custom-* id, visible only to your account. Call it exactly like any built-in model — same endpoint, same request shape, same billing path.

Next