A tier for every request
Each request goes to the efficient, fast, or frontier tier, based on what it asks for.
Built to route on purpose
Auto picks a model tier for each request, so simple work stops paying frontier prices.
Each request goes to the efficient, fast, or frontier tier, based on what it asks for.
Auto chooses between efficient, fast, and frontier tiers for each request.
Requests to debug, design, or plan multi-step work go to the frontier tier.
If classification fails, the request still runs on the default route.
Pick a model or connect your own provider, and Auto steps aside.
Where routing shows up
Auto is the default model for chat.
Runs use the same generation path as chat, so they are routed the same way.
Each tier carries its own output token cap.
Pick a specific model at any time to skip Auto.
With your own model provider connected, requests go straight to it.
Set a thinking level on models that support it.
How Auto picks a model, and how to take control.
Want help choosing models? Talk to us
Let Auto route each request, and pick a specific model whenever you want.