Model Router
Watch Jev send each prompt to the cheapest model that can handle it — fast, balanced or powerful — and count what you'd save vs. always calling the frontier. No key, just press route.
A router is one Jev call in front of your models. Get an API key →
Model prices are illustrative ($0.15 / $0.60 / $3.00 per 1M tokens, 1200-token calls) to show the shape of the savings — plug in your own to size the real gain.
The same demo is one Jev call
- Take each incoming prompt
- Ask Jev fast / balanced / powerful
- Route in parallel in front of your models
- Send each prompt to the tier Jev picked
const results = await Promise.all(items.map((item) => // items = prompts
fetch("https://jevtypesafeai.com/api/v1/decide", {
method: "POST",
headers: { Authorization: `Bearer ${process.env.JEV_API_KEY}` },
body: JSON.stringify({
state: item,
questions: {
tier: { type: "choice", instructions: "Cheapest model that can handle it?",
criteria: { fast: "small", balanced: "mid", powerful: "frontier" } },
},
}),
}).then((r) => r.json())
));
route(prompt, results[i].answers.tier.choice); // → "fast"See a real decision FAST · 198ms
Prompt: “What's the capital of France?”
How it works
Sending every prompt to a frontier model is the easy default and the expensive one. Most requests — lookups, formatting, classification, short code — are handled just as well by a small, cheap model. A router decides. Here each prompt is one Jev call that returns a typed tier — fast, balanced or powerful — plus a complexity score, all streamed in parallel, and we tally the routed cost against always calling the frontier model.
It's the same one-call-in-front-of-your-models pattern a production gateway uses, made visible: watch the split land and the savings counter climb.
Build it into your gateway
With a hosted key the same Jev call sits in front of your model pool: a typed routing decision per request at $0.25–$0.42/M input tokens, in milliseconds, returning a clean enum you can switch on directly — no parsing, no malformed output. Route the easy 80% to a cheap model, reserve the frontier for the hard 20%, and cut your inference bill without hurting quality on the requests that matter.
FAQ
Does Jev really route each prompt?
Yes — every prompt is an independent Jev call returning a typed tier and a complexity score. The batch runs in parallel and the routing decisions stream in live.
Are the cost savings real?
The shape is real; the exact numbers are illustrative. We use sample model prices ($0.15 / $0.60 / $3.00 per 1M tokens) and a fixed call size to show the saving — plug in your own prices to size the real gain.
Is it free? Can I route my own prompts?
Free, no key — a rate-limited call to the live Jev API. To route your real traffic, wire the same call into your gateway with a hosted key.
More: Context Compaction demo · PR Judge · All Jev tools · Jev API docs