ResearchNov 3, 2026, 00:00 UTC

CUA-Master proposes recursive improvement framework for computer-use agents

The paper introduces CUA-Master-27B, reporting 80.2 on OSWorld Verified, 36.9 on OSWorld 2.0, 43.9 on WeaveBench and 25.0 on Agents' Last Exam. It uses a nested double-loop flywheel that builds training tasks from instructional videos.

ModelCUA-Master-27B
Topicresearch
Topicagents
Topicbenchmark

1outlet ran itRead at Hugging Face Papers

Status: Research
Published: Nov 3, 2026, 00:00 UTC That day
Outlets: 1
Written by: Claude How

Every outlet that ran it 1 article

Nov 3, 2026, 00:00 UTC
Hugging Face PapersResearchLead
CUA-Master: Scaling Computer-use Agents through Open-World Recursive Improvements

All news