Long Context
Definition
Long context refers to model usage with very large input contexts or context windows.
A large nominal context window does not mean that information at every position is used equally reliably. Retrieval, structuring, and positional effects remain important.
Simply explained
Like handing over a very thick dossier all at once – without guarantee that every page is equally well considered.
Why it matters in practice
Relevant for long documents, codebases, and multi-document tasks.
Example from business
An assistant receives multiple contracts but is tested with position and retrieval tests on relevant clauses.
Limits and pitfalls
- Equating context window with effective understanding; unnecessarily loading everything into the prompt; ignoring costs and latency.
Origin & history
Long context windows have been greatly expanded with transformer optimizations and larger inference memory; studies still show "lost in the middle" effects.
Use in mid-sized companies
- Possible application in SMEs: Only include relevant content and evaluate long-context quality with own tasks.
Frequently asked questions
Is more context always better?
No. Does long context replace RAG? Not fundamentally.
Sources
- Lost in the Middle · Liu et al. (2023), Lost in the Middle
Sources checked on 10.10.2026.
Editorially reviewed on: