Stress-test & train frontier agents on real-world GUI workflows

CUA-Bench provides a unified, cross-platform benchmark for evaluating and training agents on realistic GUI-based work, including mouse clicks, keyboard inputs, and screen reading within dynamic desktop and mobile environments.

Developed by Snorkel's AI Data Research Lab, CUA-Bench+ delivers thousands of verifiable tasks across Linux, Windows, macOS, and Android — built to test multi-step reasoning, error recovery, and multi-application workflows.

REQUEST DATA SAMPLES //

By submitting this form, I agree to the Terms of Use and acknowledge that my information will be used in accordance with the Privacy Policy.

Task categories include

Artifact Creation
Creating a new structured professional artifact from scratch

Artifact Modification
Editing an existing artifact to meet specific structural or content constraints

Structured Export & Conversion
Exporting or converting artifacts into a specified deterministic format

Project Assembly
Constructing multi-file project structures with correct relationships and references

Cross-Tool Integration
Producing consistent artifacts across multiple applications in a coordinated workflow

Domain-Specific Structured Output
Producing structured domain-specific definitions that are internally validated by file struture

Domains include Electrical Engineering, Computer Science, Architecture, Semiconductors, Bioinformatics, Robotics, Laboratory Information Systems (LIMS), Game Development, Enterprise ERP Systems, 3D Modeling and Animation, Energy and Power Systems, GIS / Geospatial Systems, Manufacturing and Design and more.