← All benchmarks

FrontierCUA pass@1

Agents

Collinear AI's private computer-use benchmark of 15 connected workflows across running clones of enterprise applications such as Zendesk and Google Sheets; agents act only through screenshots, clicks, typing and scrolling, and an AI judge grades each attempt against a weighted rubric from the saved application state and trajectory. An attempt passes at a grade of at least 0.95, and pass@1 is the unweighted mean of per-workflow pass rates, so each workflow counts once however many attempts it has.

Top models (higher is better)