ScaleCUA-ScienceBoard-VM

# ScaleCUA — ScienceBoard VM image A QEMU/qcow2 desktop VM disk (zipped) with the six ScienceBoard applications (**ChimeraX, Celestia, KAlgebra, GrassGIS, TeXstudio, Lean**) and the evaluation REST server preinstalled. Required to run ScienceBoard evaluation of the ScaleCUA agents. - 📄 Code: [https://github.com/THUDM/SCALE-CUA](https://github.com/THUDM/SCALE-CUA) - 🤗 Collection: [https://huggingface.co/collections/extreme1228/scalecua-6a4cad5e3c80a230622eaff3](https://huggingface.co/collections/extreme1228/scalecua-6a4cad5e3c80a230622eaff3) · 🤖 [ModelScope](https://modelscope.cn/collections/extreme1228/ScaleCUA) ## Contents - `Ubuntu2_scienceboard.qcow2.zip` — ~17 GB download, ~44 GB unzipped. ## Usage ```bash pip install -U modelscope modelscope download --dataset extreme1228/ScaleCUA-ScienceBoard-VM \ Ubuntu2_scienceboard.qcow2.zip --local_dir ./vm unzip ./vm/Ubuntu2_scienceboard.qcow2.zip -d ./vm export PATH_TO_VM="$(pwd)/vm/Ubuntu2_scienceboard.qcow2" ``` Then run `osworld_eval/scripts/test_sci.sh` (see [`REPRODUCE.md`](https://github.com/THUDM/SCALE-CUA/blob/main/REPRODUCE.md)). ## Citation ```bibtex @misc{lv2026scalecua, title = {ScaleCUA: Scaling Computer Use Agents with Verifiable Task Synthesis and Efficient Online RL}, author = {Bowen Lv and Xiao Liu and Yanyu Ren and Hanyu Lai and Bohao Jing and Hanchen Zhang and Yanxiao Zhao and Shuntian Yao and Jie Tang and Yuxiao Dong}, year = {2026}, url = {https://github.com/THUDM/SCALE-CUA} } ```