This really doesn’t pan out in practice if you work a lot with these models And also we know why: effective context depends on inout and task complexity. Our b

This really doesn’t pan out in practice if you work a lot with these models And also we know why: effective context depends on inout and task complexity. Our best guess right now is that we are often between 100k to 200k effective context length for frontier, 1m NIHS type models