What it does
This experiment measures Jev as a lower-cost model router on the RouterArena benchmark.
Benchmarks & research
jev-routing-experiment
Bookmark: jev-routing-experiment
Keep this in your collection.
Leave a noteWhat would you try with this?
: jev-routing-experiment
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
A developer can adapt this code to sort questions. The app uses an outside AI service called TypeSafe to check how hard a question is. Then, it sends the question to the cheapest AI that can answer it. To connect them, the developer needs a special service code from TypeSafe, not a personal password.
The test showed that just sorting by difficulty did not beat using one smart AI. The best method was finding similar past questions first. A developer can start by looking at the small practice test in the source files. This shows exactly how the sorting rules work.