Cursor's usage limits are primarily reflected in the number of fast requests—i.e., the number of times high-performance models are used for code generation and conversation.