I think 200k context window w/ good compaction is the sweet spot. Models claiming to have larger are rarely better at rembering important stuff from my experience↗·@jullerino·Apr 21, 2026·essay·context-window·memory·model-performance
@yiannis__p yea github.com/pingdotgg/t3co…↗·@jullerino·Apr 16, 2026·tool·expo·github·mobile-client·t3code