- Who it helps
- Researchers evaluating training data for computer-using agents.
- Availability
- November 2 paper and implementation are public. The linked model card was inaccessible during this check.
How examples are built
Reusable task steps become verifiable tasks. Restoring a saved desktop state lets the system retry actions and keep successful sequences of actions for training.
CUA-Master generates and verifies computer-use training tasks ↗
What reproduction requires
The repository requires model-provider access and a real virtual-machine or sandbox backend. Its mock runner does not establish successful interaction with a real desktop.
CUA-Master generates and verifies computer-use training tasks ↗CUA-Master implementation ↗CUA-Master author page ↗
What you can do
- Inspect the backend and evaluation requirements. Open reference ↗
Where the evidence stops
Paper and repository checked November 3. Benchmark gains are author-reported; no training, inference or desktop workflow executed. Provider costs and infrastructure remain prerequisites.
Sources read. Primary sources read November 3, 2026 UTC. No third-party software, model inference or full workflow executed.
Sources & further reading
Paper · paper
CUA-Master generates and verifies computer-use training tasks ↗
Identifies the subject, documented capability or specific verification limit.
arxiv.org · page checked 2026-11-03
GitHub · implementation
CUA-Master implementation ↗
Identifies the subject, documented capability or specific verification limit.
xlyu0106 · page checked 2026-11-03
Hugging Face · paper
CUA-Master author page ↗
Identifies the subject, documented capability or specific verification limit.
CUA-Master authors · page checked 2026-11-03
Coverage note: Independent attention and end-to-end execution remain unverified.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
CUA-Master builds computer-use training examples from workflows mined from videos.