What is Context rot?
Also called: Context degradation
Context rot means language models work less precisely the more text sits in their context window. They miss details, mix things up or follow instructions less closely.
Accuracy
Example
After a long chat with many files, an assistant forgets an instruction from the beginning. In a new chat with a short summary it works again.
For pros
A technical report by Chroma showed for many tested models that reliability drops as input length grows, even on simple tasks. A related finding is “Lost in the Middle”: information in the middle of long inputs is used less well. Remedies from context engineering: put only relevant content into the context, summarise long histories (compaction), retrieve selectively with RAG and use subagents with their own context. A large context window does not mean every detail is used equally well.
Related prompts, skills, agents and tools
Written by AI Compare Ninja with AI assistance and edited by us. Every explanation is our own text.