AsideAI cuts compaction/dreaming token use 7x, doubles cache hit rate

garrytan · x · 2026-09-29

AsideAI announced a round of token-efficiency tuning (retweeted by Garry Tan): compaction and dreaming now use on average 7x fewer tokens (up to 14x), cache hit rate improved 2x, and accuracy increased rather than dropped, running on Fable 5.1. The team says new harness updates shipping soon will further improve token efficiency and speed.

Related event: AsideAI Cuts Token Consumption Up to 14x with Compaction and Dreaming(2 posts)→

Original post →

More from coding & agent

coding & agent channel →