Efficiency when returning to a session
retained text
Hi all. Sometimes i let my context grow to 800k (i know naughty me), and then i hit my limits and return to a cold cache of 800k tokens. Is it more efficient to use haiku to run /compact and then switch to opus? I know that the cache status doesn't transfer between models. Has anyone tested this? Side note on the topic of compacting and where to set the auto compact, has anyone tested an optimal setting for this in terms of efficiency? Obviously setting auto compact to 80k tokens would waste so many tokens on the compacting process itself, as well as rereading new info and files, so is there a magic number i should be using? Thanks!
document fetched 2026-10-26 00:37 · canonical https://reddit.com/r/ClaudeCode/comments/1xeu0oc/efficiency_when_returning_to_a_session