In early December last year, the OpenAGI Foundation made a big move—they launched a foundational model called Lux. This thing is pretty interesting, claiming to be the first open-source solution specifically designed to teach AI to "click a mouse and type on a keyboard like a human."

Simply put, in the past, no matter how smart AI was, it still had to work through API interfaces. Now, Lux can directly operate software interfaces. They tested it with 300 everyday task scenarios, and the results were quite impressive: it scored 83.6% on Online-Mind2Web, a widely used industry benchmark.

To put that into perspective—a major search giant’s Gemini CUA scored 69%, a certain chatbot company’s Operator got 61.3%, and even Claude Sonnet, a product from a company focused on AI assistants, didn’t surpass this score.

The open-source approach really has something going for it, at the very least lowering the barrier for more developers to get hands-on with “AI automated computer operations.”