CUA-Lite: An Open Platform for Computer Use Agents

Key points

• Open resources for computer use agents (CUAs) -- environments, datasets, and models -- remain scattered: without a standard, these resources cannot be integrated and scaled; additionally, evaluation, RL, and SFT cannot share one infrastructure.1
• (2) Unified supervised data format, into which we convert the major public CUA datasets, from GUI grounding and understanding to full rollouts, and record our fresh rollouts from frontier CUAs on these verifiable environments.1
• We release reference scores across the integrated benchmarks, and show that RL and SFT through the same harness improve open-weight models.1
• We release our dataset, code, and models, and the platform is open to contributions.1

Sources (1)

• [1]CUA-Lite: An Open Platform for Computer Use Agents
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 09:30 AM

  We present CUA-Lite, an open platform for developing CUAs, built around three standardized abstractions: (1) Unified environment interface, under which we integrate 12 benchmarks across desktop, browser, and mobile, plus lightweight VM-free sandboxes that replicate OSWorld's desktop at much lower cost, providing over 30,000 verifiable tasks.

  Open resources for computer use agents (CUAs) -- environments, datasets, and models -- remain scattered: without a standard, these resources cannot be integrated and scaled; additionally, evaluation, RL, and SFT cannot share one infrastructure.

Extractive summary: sentences quoted from the sources.