The right model
for every task.

Cheapest eligible model, task by task.

Describe each task — classify, summarize, extract, draft, code or reasoning — with monthly requests, average input and output tokens, the quality it needs, how fast it must answer and how sensitive its data is. Enter the models you can use with their per-million-token rates. The router applies four rules, picks the cheapest model that passes all of them, and explains every choice.

Practice on fiction

Sample: a support + ops workload.

A fictional outfitter's six tasks and five generically named models with clearly labelled example rates — so the routing, savings and exports work before you enter your own provider pricing.

⇢

Four routing rules

Quality tier at or above the need; a model fast enough for real-time or interactive work; cleared by your policy for the task's data; and a context window that fits.

%

Savings and what-ifs

Routed monthly cost against everything on the top model, plus cache-hit and batch-discount sliders and a budget cap that warns before you overspend.

⇪

Policy, CSV, route config

Export a Markdown routing policy, a formula-safe CSV of routes and costs, and a JSON route config your engineers can wire into a gateway.