More Thinking Tokens May Cause Context Rot

dejavucoder · x · 2026-07-13

The author observed that in some cases, high mode performs better than xhigh mode, or at least similarly, but with lower costs and faster speeds.

The speculated reason is that more thinking tokens might actually lead to worse context rot, meaning the thinking process scrambles the context the more it ponders.

Original post →

More from Models

Models channel →