Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

underproofed ganbold@ganbold (2026-09-17 10:24:26)
frontier models are pain in the head for QnA compared to their flash version

I prefer a pair of flash and frontier, but currently there is no harness that could handle this kind of workflow.

you ask flash => flash decide to answer directly or delegate it to frontier => you got answer from flash

I don't think its just a matter of that frontier model is smart or token inefficient for such a task, but rather like I think that is how it behaves