# Built-in models, quota and windows

Pix Smart and Pix Fast, the models your plan includes, how calls are routed behind each, your allowance and its windows, and which models each host can use.

Your plan includes two model **labels**:

| Label | For |
|---|---|
| Pix Smart (`pix-smart`) | Harder work: coding, planning, review, images |
| Pix Fast (`pix-fast`) | Quick work |

You only ever see the label. Behind each, Vulnesis ranks up to five Workers AI models. Every call is routed across them by recent performance: answers, refusals and speed. If the first is struggling, the call goes to the next. Model ids never appear in the Resolve Platform, sessions or errors.

## Allowance and windows

Each tenant has an allowance of built-in tokens a month. The window is the UTC calendar month: it starts at 00:00 UTC on the 1st and resets at 00:00 UTC on the 1st of the next month. Refused and failed calls do not count. When the allowance runs out, the factory does what `onAllowanceExhausted` says: fall back to your own keys, pause, or buy a top-up. See [Model routing](/software-factory/cost/model-routing.md).

The factory's share of the allowance is `budget.factory_share` (75%), and `pix.reservePercent` (20%) is kept for your own sessions.

## Hosts

The Belai build in a Pix Sandbox carries a built-in provider that offers Pix Smart and Pix Fast, so its sessions can pick either like any provider. Pix Smart takes image input; Pix Fast does not.

## Related

- [Model routing](/software-factory/cost/model-routing.md)
- [Usage and spend](/software-factory/cost/usage.md)
