Send every prompt to the cheapest model that can answer it.

Sluice reads each request, picks the least expensive model that can handle it, checks the answer, and steps up a tier only when it has to.

Top modelanthropic/claude-fable-5.1

$0.015210

Routedanthropic/claude-haiku-5.5

$0.000152

You save $0.015058 (99.0%) on this prompt.

Estimate at list price for ~21 input and 300 output tokens. Classified as easy. Run it for real in the playground to see actual costs.

Two-line migration

Keep your OpenAI client. Change the base URL and the key.

TypeScript

const client = new OpenAI({
  baseURL: "https://sluice.forum/api/v1",  // was https://api.openai.com/v1
  apiKey: process.env.SLUICE_KEY,  // model: "auto" lets Sluice pick
});

Create a key with a spending limit in the dashboard, under Gates.

How routing works

1
Classify
Rules read length, code, math and the kind of ask, and sort the request into easy, medium or hard. No model call.
2
Pick
The cheapest model in that tier that exists in the live catalogue and fits the request.
3
Check
Empty, refused, cut-off or broken-JSON answers are retried once on the next tier up.

Prompts are never stored.