arXivPranjal Aggarwal, Lawrence Keunho Jang, Sean Welleck, Daniel Fried, Ruslan Salakhutdinov, Jing Yu KohWed, Sep 30, 2026, 10:48 AM PDT
score 17.1
New Benchmark Measures How Fast Computer-Using AI Agents Really Are
Original: cua-speedrun: Standardized Benchmarking of the Speed of Computer-Use Agents
Source: arxiv.org ↗
Writing ELI5 summary…