Context Rot: How Increasing Input Tokens Impacts LLM Performance
by Chroma
Measures how output quality drops as input length grows.
The research investigates how the amount of input changes model performance. It encourages measuring the effect of extra context instead of assuming that a larger supported window means every detail is used equally well.
- Level
- Advanced
- Models
- any
- Category
- Agents and harnesses
- Tags
- context-engineering, research
- Source date
- date not stated
- Added
- 2026-10-18
- Last verified
- 2026-10-18
- Status
- ok
Summary in our own words. All rights with the original creator. Correction or removal: DM on X.