# Context Rot

> Context rot is the decline in a model's reliability as the amount of input in its context grows, even on simple tasks and well within its advertised context window.

- Identifier: PTL-0169
- Category: Failure Modes & Evaluation
- Canonical URL: https://protologue.com/t/context-rot/
- Also known as: long-context degradation, context degradation
- Introduced: 2025

## Description

A 2025 study of 18 models found performance degrading with input length in ways that depended on the task and on the content around the relevant information, such as similar-looking distractors. It is a practical reason to keep contexts lean, retrieve only what is needed, and compact long histories rather than relying on maximum context length.

## Related terms

- [Lost in the Middle](https://protologue.com/t/lost-in-the-middle/)
- [Needle in a Haystack](https://protologue.com/t/needle-in-a-haystack/)
- [Context Window](https://protologue.com/t/context-window/)
- [Context Engineering](https://protologue.com/t/context-engineering/)
- [Context Compaction](https://protologue.com/t/context-compaction/)

## Sources

- Hong et al. (2025). Context Rot: How Increasing Input Tokens Impacts LLM Performance. https://www.trychroma.com/research/context-rot

## Cite this entry

Protologue. (2026). Context Rot. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0169). https://protologue.com/t/context-rot/

License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)