Context Window
The maximum span of tokens or model state that a language model can directly consider while generating or scoring a sequence.
Language-Model Context
The context window is the amount of tokenized state that a language model can directly condition on during scoring or generation. Prompt text, conversation history, tool results, retrieved material, and generated tokens can all consume the same budget.
Capacity Boundary
A model's advertised maximum context length does not mean information is used equally well at every position. Long-context quality, positional behavior, KV-cache memory, latency, and retrieval strategy must be evaluated separately.