SUPER INTELLIGENCE ROUTER

The right model
for every request.

One virtual model name. ZRouter reads each request, judges how hard it is, and sends it to the cheapest model that can handle it. Your app never changes.

Why it matters

• Lower cost. Most traffic is greetings, lookups, rewrites, and short answers. Those go to a fast, low-priced model instead of your most expensive one.
• Better answers where it counts. Debugging, multi-step reasoning, long documents, and high reasoning-effort requests go to your strongest model.
• No code changes. Your app keeps calling the same model name. You decide which providers serve each tier, and change it any time.

Set it up

1. Open Virtual models in your dashboard and create a virtual model.
2. Add the targets you want, for example a fast model, an everyday model, and your strongest model.
3. Choose Super Intelligence Router as the strategy.
4. Tick which targets serve simple, standard, and complex requests. You need at least one tier.
5. Send requests to the virtual model's selector, such as accounts/<account-id>/auto.

You can also set it up from zctl or ask your AI assistant through MCP.

How it decides

Each request gets a difficulty score from 0 to 100, in well under a millisecond, from what is in the request itself:

Below 25 is simple, 25 to 54 is standard, and 55 or more is complex. If a tier has no available target, the request moves up to a stronger tier first, so a hard request is never sent to a weaker model while a stronger one is up.

See every decision

Each response includes a header that explains the choice:

X-ZRouter-Route: super_intelligence; tier=complex; score=63; reasons=code,reasoning-cues

The model that served the request also appears in your request logs and usage, so you can check the savings.