Context rot
Definition
The quality drop that happens as you give a model more context. A million-token window doesn't use all of it equally well — accuracy falls, instructions in the middle get ignored, and the model leans on the wrong parts. It degrades gradually, so nobody notices the day it started.
Context rot is the quality drop that happens as you give more details to the LLM. A model with a million-token window does not use all of it equally well: accuracy falls, instructions in the middle get ignored, and the model starts leaning on the wrong parts of what you gave it. It degrades gradually rather than failing outright, so nobody notices the exact day it started.
Why it matters
Context rot breaks the instinct everyone has, which is to fix a bad output by adding more — more examples, more rules, more retrieved documents, longer chat history. Past a point, each addition makes things worse. Long-running agents and chat features hit this hardest: the session that works fine at turn 3 is quietly unreliable by turn 40. The only way to see it is to score the same cases at different context lengths.
Related terms