I mentioned this at work but context still rots at the same rate. 90k tokens consumed has just as bad results in 100k context window or 1M. Personally, I’m on

I mentioned this at work but context still rots at the same rate. 90k tokens consumed has just as bad results in 100k context window or 1M. Personally, I’m on a 6M+ like codebase and had no problems with the old window.