HOW WE COMPARE MODEL QUALITY IN CURSOR

Summary from the TLDR AI newsletter

Cursor uses a hybrid online-offline eval process to keep its understanding of model quality aligned with what developers actually do. The offline part uses CursorBench, an internal eval suite based on real Cursor sessions from Cursor's engineering team. It measures multiple dimensions of agent performance, including collusion correctness, code quality, efficiency, and interaction behavior. The online part involves a controlled analysis on live traffic, which catches regressions that offline suites miss. Together, this loop keeps Cursor's notion of model quality grounded in production as workflows change.