The company OpenAGI debuts in the market with an announcement that could destabilize the current market of artificial intelligence services. It will launch Lux, a new “agent” AI model capable of controlling computers more efficiently than established systems developed by giants OpenAI and Anthropic.
Founded by Zengyi Qin, an experienced researcher at MIT (Massachusetts Institute of Technology), the company takes a different approach in the race for automation. While the big players have focused on generalist language models, OpenAGI is betting everything on Lux, which is a fundamental model designed specifically for autonomous action and execution in the digital environment.
This is not just another chatbot. Lux was built to operate computers autonomously, mimicking the behavior of a human user. The technology behind Lux allows it to “see” the screen by interpreting screenshots and perform complex actions in various desktop applications. This ability to understand and act on the GUI is a big step forward for AI agents, transforming AI from a passive consultant to an active operator.
The company’s promise is twofold: superior performance and low costs. In an industry where computing power is extremely expensive, the claim that Lux can outperform market leaders at a fraction of the price is an extremely strong business case for companies looking to automate complex processes.
To back up its bold claims, OpenAGI has put concrete results from a respected benchmark on the table. According to the data provided by the company, the Lux model achieved a success rate of 83.6% in the Online-Mind2Web test. This benchmark has become the de facto industry standard and is considered the most rigorous test for evaluating artificial intelligence agents specializing in computer control.
Achieving such a high score puts the startup in a favorable position, suggesting that their technology is mature enough to handle real tasks, not just controlled demos. Success in this test indicates an advanced ability to navigate menus, fill out forms, and interact with dynamic web elements, overcoming the limitations of previous models that often bogged down in the face of complex or unfamiliar interfaces.
Going public with this announcement marks the beginning of a new phase for OpenAGI, which now competes directly with the virtually unlimited resources of OpenAI and Anthropic. It remains to be seen how the market will react and whether the performances in the tests will translate as well in the everyday use scenarios of companies.
