CoworkOS harness scales agent coworkers for real-world automation
Original title: CoworkOS:Scaling Agent Coworkers for Real-world Automation
CoworkOS uses GLM-5.3 backbone to post-train Qwen3.6-35B-A3B, raising average benchmark scores by 9.76 points.
CoworkOS separates task strategy from execution management in a three-layer architecture for long-horizon work. It supports hierarchical workspaces, persistent semantic memory and human-in-the-loop inspection of shared documents. Evaluated across 12 benchmarks and 6 real-world application categories, it competes with frontier agents such as Codex.
- 9.76 point improvement after five post-training rounds
- Uses GLM-5.3 as backbone model
- Hierarchical workspaces ensure continuity
- Direct human modification of artifacts possible
From the original article
1 min read at huggingface.co
Rapid advances in cowork agents are shifting human-AI collaboration toward real-world automation, delivering software, designs, and other artifacts. However, such productivity work demands both high-quality delivery on complex productivity tasks and continued monitoring and decision-making over long horizons, a combination that remains an open challenge.
By Jiajie Jin, Guanting Dong, Chengjun Pan · 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.