Can a router answer as well as always using the big model, for less?
Send each request to the cheapest model that still gets it right.
Best in this use case · ranked only on systems that ran every core taskopus-onlymodelGot 91% of cases right, averaged over 1 core task.
Best value · at least 10% cheaper, still at least ¾ of the best scorejev-cascade-3085% right · $1.30 a run · 45% cheaperCheapest that works: haiku-only · 67% · $0.50
bar = share of cases right · each row is one full system
⚡ Jev as a router: cheaper than always-Opus?model-routing200 multiple-choice questions from MMLU-Pro — 100 law, 100 engineering — where cheap and frontier models are far apart. A solution answers each question however it likes: one model, or a pipeline that sends some questions to a cheaper model and some to a stronger one. The board p200 cases · 6 runs · core taskOpen task: rules, cases, every run →