Comment by mrtesthah
17 hours ago
Most models’ reasoning abilities drops off significantly between the 256K-1M token ranges of the context window. There’s too much stuff to “pay attention to” at once.
17 hours ago
Most models’ reasoning abilities drops off significantly between the 256K-1M token ranges of the context window. There’s too much stuff to “pay attention to” at once.
I auto compact around 200k tokens both due to this and because the cached read cost really escalates when sessions have more tokens than that (too short and you pay a lot in per-compact re-reading of state)