Comment by ravroid

I imagine this would be cost prohibitive at scale since it would require two models to run for every user message?