Updates
1. CursorBench 4.0 adds longer multi-file coding tasks
The September 10 changelog adds long-horizon tasks covering edits, refactors, investigation, intent understanding, job management and design adherence.
Proof: The source records 4.0 methodology.
Impact: Teams comparing coding agents can distinguish performance on ambiguous multi-file work from short standalone coding questions.
Watch next: For Cursor, recheck methodology and model pricing when comparing a later benchmark update.
Canonical host: cursor.com