Cua Open-Sources 4B Multimodal Model Cua-S1-4B-0.2
A 4B model hitting 92.9% on GUI automation means agents that control a computer need no heavy infrastructure and can run on modest hardware, making open-source computer-use agents far more practical. Developers building desktop automation or agent frameworks can grab the weights from the model link and fine-tune for their own workflows. It competes with much larger closed computer-use offerings from US labs at a fraction of the footprint.
Cua has released Cua-S1-4B-0.2, an open-source 4B-parameter multimodal model that performs computer operations by reading screen visuals and task goals directly.
The model was trained with reinforcement learning using task-completion rewards in real system environments, without relying on accessibility structures beneath the software.
On a GUI benchmark of 168 tasks, it achieved a 92.9% task completion rate, well above the 60.
1% of the untrained baseline.
Cua开源多模态操作模型：轻量模型辅助决策
开源4B参数模型Cua-S1-4B-0.2，专门通过识别屏幕画面和任务目标来直接执行电脑操作。该模型在真实系统环境中通过任务完成奖励进行强化学习训练，无需依赖软件底层的无障碍结构。在168项GUI基准测试中，其任务完成率达到92.9%，大幅超过未训练基线模型的60.1%。
模型：https://t.co/3WOA7fuouX
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Cua, Cua-S1-4B-0.2, GUI automation, open-source models.
Source attribution: curated from Gorden Sun (@Gorden_Sun).
