latency and compute comparisons highly depends on your local setup.

you can swith to a better model for lower error rate.

Which massively slows down the output. Doing this with Qwen 9B already takes you into seconds per answer territory, and Jev is supposedly frontier level intelligence.