Back to Glossary
Context Window
Models & Architectures
Maximum number of tokens a model can process at once.
The context window defines how many tokens from the input or conversation history a model can attend to.
- Effects: Determines memory span and output quality in long tasks.
- Trade-offs: Memory/compute vs. range; long-context methods include sliding windows and sparse attention.
- Practice: Chunking, retrieval (RAG), summarization, structured prompts.