The models have reached, on average, the 256k/1M window size and that is more than enough for them to complete most tasks without losing attention. I suspect more people are using fresh sessions and markdown between them, than trying to keep fewer long running sessions - this works better overall. We are moving more into the quality >>> quantity realm.