Microsoft has developed Mage-VL-Cowork, a 4B small vision-language model designed for autonomous desktop agents. This model achieves top performance at its scale, scoring 66.0% on the OSWorld-Verified benchmark. It utilizes an efficient curriculum and workflow synthesis to master hybrid computer tasks, requiring fewer model calls and output tokens. AI
IMPACT Sets a new benchmark for small vision-language models in autonomous desktop agent capabilities.
RANK_REASON Frontier-lab model release with system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →