DictionaryContext Management
Knowledge Cutoff
The date when an AI's learning stopped; it does not know about anything newer unless you tell it.
Definition
The date after which a model has little or no training knowledge, so newer libraries, APIs, and events must be supplied in context. Anthropic lists two dates per model: the reliable knowledge cutoff (through which knowledge is most extensive and reliable) and the training data cutoff (the broader range of data used). Claude Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 list June 2026 for both.
Example
A model with a June 2026 cutoff will not know a framework API released in September 2026, so paste the changelog or let it fetch the docs.