Hangzhou AI agent tops global benchmark with record 90.2% success rate
A Hangzhou-based artificial intelligence company has topped the latest global OSWorld benchmark, with its AI agent achieving a record-breaking 90.2 percent task success rate.
Developed by Intelligence Indeed, the agent ranked first overall and also led the Agentic Framework category. In doing so, it has become the first AI agent to surpass the 90 percent mark since the benchmark was introduced in 2024.
Created by researchers from the University of Hong Kong, Carnegie Mellon University in Pittsburgh, Pennsylvania, and the University of Waterloo in Ontario, Canada, OSWorld is widely regarded as a leading benchmark for evaluating AI agents' ability to operate computers.
It tests more than 300 real-world tasks, including document editing, spreadsheet processing, image manipulation, email management, and cross-application workflows.
The latest score marks a significant leap in AI performance. Top models achieved only about 12 percent success when OSWorld debuted in 2024, while the previous record hit 83.6 percent this May.
Founded in Hangzhou, Zhejiang province, in 2018, Intelligence Indeed has provided AI automation solutions to more than 6,000 enterprises, including Fortune Global 500 companies, State-owned enterprises, and industry leaders across sectors such as finance, manufacturing, energy, and transportation.
The company said the result reflects not only the advances made in large language models but also improvements in engineering systems that enable AI agents to reliably complete complex, multi-step tasks, bringing autonomous "digital employees" closer to real-world deployment.




