Lotte Innovate, the IT services arm of Lotte Group, said Monday it has upgraded the language model powering its artificial intelligence platform iMember.
The upgraded model has been enhanced to provide in-depth explanations and logical responses to user queries. It shows particular strength in organizing context and constructing answers even for complex questions. To achieve this, the company significantly expanded the model's parameters from 32 billion to 120 billion.
The model also posted strong results in objective performance benchmarks. It scored 78.31 on the "Suneung Korean QA" test based on Korea's national college entrance exam Korean language section, followed by 78.15 on the "Haerye Bench" evaluating comprehensive Korean-language capabilities, 60.67 on "Expert Knowledge Reasoning," and 92.98 on "Mathematical Reasoning."
Lotte Innovate's model is an open-source-based, on-premise model that operates on the company's own infrastructure without relying on external APIs. This satisfies data security, compliance, and customization flexibility — all considered critical requirements in enterprise environments. Beyond performance, speed and efficiency have also improved. The system was optimized using the latest model architecture with quantization technology, boosting inference speed approximately threefold under the same GPU resource conditions. This allows the company to stably run the 120 billion-parameter large model using infrastructure originally built for the 32 billion-parameter model, while also improving user-perceived response times and significantly enhancing real-time business application.
The platform, previously centered on Korean, has added English and Japanese language support. Going beyond simple translation, the company expanded the UI/UX to allow users to directly operate and navigate the platform in those languages. This lays the groundwork for smooth adoption of iMember across Lotte's overseas subsidiaries and affiliates.
"We will continue to upgrade iMember's performance, speed, and security levels, while steadily advancing industry- and task-specific models to lead AI innovation for enterprises," a Lotte Innovate official said.