Korea's Enhans makes global top 3 in AI agent benchmark

(Enhans)

Enhans, a South Korean startup specializing in vertical commerce AI agents, said Tuesday that its artificial intelligence model, ACT-1, ranked third in the global benchmark for web-based AI agents, Online Mind2Web.

The model placed just behind global tech giants OpenAI and Anthropic.

The Online Mind2Web benchmark assesses AI agents on their ability to perform complex tasks on the web, emulating human behavior. The evaluation covers 300 real-world tasks across 136 actively operating websites, assessing how effectively AI agents can navigate and interact with diverse web environments.

ACT-1 is a behavior-based AI model tailored for web automation. It can execute sophisticated actions such as clicking buttons and navigating menus.

The company emphasized that, unlike conventional approaches that rely on screenshots and coordinate-based controls, ACT-1 employs a document object model-based method to directly access and manipulate webpage structures.

Enhans highlighted its No. 1 ranking in the commerce domain among global competitors, underscoring the potential of Korean-developed AI agents on the world stage.

Building on this momentum, the company plans to evolve ACT-1 into a next-generation agent by integrating imitation learning, enabling it to perform tasks traditionally limited to human users.

Additionally, Enhans intends to expand its enterprise-level web automation solutions, seeking to differentiate itself from global AI firms and secure a unique competitive edge.

“Being able to stand alongside global AI leaders like OpenAI and Anthropic is profoundly meaningful. We will continue to strengthen the development of our web AI agent technologies to lead the global market in vertical industries,” Enhans CEO Lee Seung-hyun said.