Model selection best practice

I noticed that when using Sonnet, tasks usually take longer and usage limits are spent slower, but looks like opus does the same tasks faster, so i't incorrect to compare them by just limit usage/minute, but rather by limit usage/task.

Does anyone have some practical understanding of what the tradeoffs look like?

I was also thinking of having smth like model selection feature, where GSD can autonomously assign a model depending on task type and complexity to save usage where possible and not sacrifice with quality where required. Or maybe this feature already exists and I just don't know about it.

0 comments