Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2610.12611.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

SoftwareGym is an automated pipeline that synthesizes 182 diverse GUI applications across twelve categories, with runtime-verified functionality via GUI evaluation and iterative repair. The authors build SoftwareGym-Bench (120 tasks) and use collected trajectories to train a computer-use agent, improving Pass@3 from 41.39% to 56.97% on the out-of-domain OSWorld benchmark.

Concrete framework and benchmark for scaling computer-use agent training through synthesized environments, with reported transfer gains on OSWorld.

Cited by

No citations on record.