> Roughly 800ms per decision with 5 questions in one call, versus 10-30s of LLM round-trips before.

> Roughly 800ms per decision with 5 questions in one call, versus 10-30s of LLM round-trips before. That's useful... I need that kind of speed, I've tried the smaller models that have lower latency on openrouter, and the ones tied with cerebras, or groq, but those are all unstable... really hoping for jev to be better