Group Pinning

Every model is served by one or more provider groups, each with its own price. By default a key uses auto: every request routes to the cheapest group that is currently live.

Pinning overrides this per model: you pick exactly which groups it may use, and billing is locked to their rates. Useful when you need a specific provider's quality or region, or a rate that must not drift.

Each model's detail page lists every group with its input and output price under Group Pricing; the discount badge in the model list always reflects the cheapest group.

Open API Keys, create or edit a key, and open the group selector. Search for a model, click it, and tick one or more provider groups, sorted cheapest first with their rate multiplier. Models you leave alone stay on auto.

Pinned models float to the top with a badge showing how many groups they are locked to, and Clear all resets the key to full auto. The key list shows pinned keys as auto +N.

A pinned model routes only within its pinned groups: the cheapest is tried first with automatic failover to the next on errors or downtime. Whatever group actually serves the request is the rate that gets billed.

An explicit X-Group header on a request still wins over the pin, and unpinned models keep cheapest-available routing.

Pair a pin with a watch to hear when its groups change price or go down, see Notifications

If every pinned group for a model is down while other providers still serve it, requests fail with a targeted error instead of a generic busy message:

{
  "error": {
    "message": "Your pinned provider groups for model \"glm-5.2\" are currently unavailable, but the model is still served by other providers. Edit this key's group override, or try again later.",
    "type": "new_api_error",
    "code": "get_channel_failed"
  }
}

Widen or remove the pin on the key, or retry later. The model itself is still up.