Bench AI agents on real computer-use tasks: AutoCAD, Xcode, Figma, Docker
Grader Labs publishes Corpus 01, a public sample dataset of computer-use (CUA) tasks: 8 full screen recordings with frame-level action trajectories and verifiable graded outcomes across AutoCAD, Xcode, Figma, Docker Desktop, Miro, and more. Sized for one sitting, it benchmarks whether agents can operate real professional tools, not demos.