Jev cuts model-routing latency 95% and cost 5x, one test finds

Ivan Zhou @ivanzhouyqOriginally posted 2026.09.17

Open original on X

I tried @typesafeai’s Jev on a few model-routing tasks and was really impressed by its speed: 95% lower latency than GPT-5.6 Luna, at 5× lower cost. It is fast enough to run inline during an agent session, monitoring every turn and dynamically switching models when needed. Such latency unlocks a new class of real-time, adaptive agent experiences!