Hugging Face paper introduces NorthCUA for scaling computer-use agents
Original title: NorthCUA Technical Report: Turbocharging Computer-Use Agents through Experience at Scale
NorthCUA turbocharges computer-use agents through an experience flywheel and staged learning achieving state-of-the-art results among open-source models.
Scaling computer-use agents requires connecting decisions to outcomes via experience. NorthCUA integrates an Experience Flywheel supplying executable tasks to a Harness for long interactions and outcome checks. The Flywheel curates verified trajectories and uses recursive self-improvement for unsolved tasks. Staged Learning combines experience with grounding data for GUI competence through mid-training and long-horizon fine-tuning. The resulting Nano Mini and Pro models achieve state-of-the-art results in GUI grounding desktop execution and Web interaction. NorthCUA-Ultrafast pairs direct selection with open-ended generation for responsive workflows.
- Experience Flywheel supplies tasks to Harness
- recursive self-improvement for unsolved tasks
- state-of-the-art among open-source models
- Nano Mini Pro family
- direct action selection with generation
From the original article
1 min read at huggingface.co
Scaling computer-use agents from individual actions to complex software work requires experience that connects decisions to complete outcomes. We present NorthCUA, an integrated system for turbocharging computer-use agents through experience at scale. Its Experience Flywheel supplies diverse executable tasks to the NorthCUA Harness, which sustains long interactions, records decisions, and checks task outcomes.
By Chaoyun Zhang, Bingxuan Li, Pinxue Guo · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.