Long Context
Using a model's ability to process very large amounts of text in one go, instead of breaking it into chunks and searching. Claude Fable 5, Opus 5 and Sonnet 5 each handle up to 1,000,000 tokens — roughly 555,000 words — in a single context window. For many use cases (analysing a full legal contract, reading a codebase, processing a set of transcripts) it is simpler and more accurate to give Claude the whole thing than to build a search system around it.
In practice
You paste all 200 pages of a vendor contract into Claude and ask "what are the termination clauses and auto-renewal terms?" Claude holds the entire document in one session rather than chunking it and losing coherence. The tradeoff is cost: everything in the window is billed on every call, so for high-volume automated work, retrieval can still be cheaper than long context even when long context would work.
Related concepts