Trillion Labs, an artificial intelligence model startup, said on Sunday it has developed gWorld-32B, a mobile world model that learns causal relationships in surrounding environments to simulate future changes.

gWorld-32B is a technology that reproduces the next screen state in real time based on a user's touch input in a mobile environment. Unlike existing world models that generate the next screen at the image (pixel) level, causing text blurring or shape distortion, this model predicts in the form of executable web code (HTML and CSS).

The company said this lowered the rendering failure rate to less than 1 percent and secured high accuracy that keeps text and icons clear without distortion. It also recorded graphical user interface (GUI) prediction performance that exceeds that of the ultra-large AI model Llama-4-402B, which has more than 50 times as many parameters.

The model delivered zero-shot performance on the Korean-language benchmark Korea Mobile App Manipulation Performance Evaluation Index (KApps), carrying out tasks immediately without additional training. This shows it is optimised for Korea's app ecosystem and UI structure compared with global general-purpose models, the company said.

Trillion Labs plans to deploy the technology immediately in industrial settings including mobile agents, automated app testing and automated customer response.

Jae-min Shin (신재민), chief executive of Trillion Labs, said, "Beyond question-and-answer AI, the era of 'actionable AI' that acts directly in complex environments has arrived." He added, "We expect it will enhance South Korea's AI sovereignty and industrial competitiveness across a range of areas including robotics and public service automation."